#387 – George Hotz: Tiny Corp, Twitter, AI Safety, Self-Driving, GPT, AGI & God
03:15:19 Share

The following is a conversation with George hots, his third time in this podcast. He's the founder of coma ai six of time was driving and is the founder of a new company called tiny corp. That created tiny grad, a neural network framework that is extremely simple, with the goal of making IT run an a device by any human, easily and efficient.

As you know, George also did a large number of fun and amazing things, from hacking and iphone to recently joining twitter for a bit as an intern in quotes, making the case for a reactor, the twitter code base in general. He's a fascinating engineer and human being, and one of my favorite people to talk to. And now a quick few second mention of the sponsor.

Check them out in the description is the best way to support this podcast. Weve got numerous for the world's hardest data. Cites ternant bible for learning new languages. That sweet for business management software. Inside tracker for blood paneling and ag one for my daily multivitamin ese wise than my friends.

Also, if you want to work on our team, we're always hiring, got the next three ment dark com slash hiring and now I want to, the flight reads, as always, no ads in the middle. I try to make this interesting, but if you must keep them friends, please still check out response, or I enjoy their stuff. Maybe you will do this.

Epo de has brought you by numerous a hatch fund that uses artificial intelligence and machine learning to make investment decisions. They created a tournament that chAllenges data scientists to build best predictive models. Financial markets.

It's basically just a really, really difficult real world dataset that test out your ideas for how to build machine learning models. I think this is a great education platform. I think this is a great way to explore, to learn obama learning to really test yourself on real world data with consequences.

No financial background is needed. The models of scored based on how will they perform an unseen data, and the top performers receive a share of determined prize pool. Head over to numerous slash legs as N U M E R dot A I slash legs to sign up for tournament and hold your machine learning skills that new more slash looks for chance to play against me and win a share of determined Price pool that's num flash legs.

This shows also brought to buy bible and APP and website to get you speaking in a new language. For ten weeks, I have been using IT to learn a few languages, spanish, a, to review russian, to practice russian, to revisit russian from a different perspective, because that becomes more more relevant for some the previous conversations we've had in some upcoming conversations I have. IT really is fascinating how much another language, knowing another language, even to degree, where you can just have little bits and pieces of a conversation, can really unlock and experience in another part of the world.

When you travel in france, in paris, just having a few words, your disposal, a few phrases, IT begins to really open you up to strange, fascinating new experiences that ultimately, at least to me, teach me that we're all the same. We have to first see our differences, to realize those differences are grounded in the basic humanity. And that experience that we're all very different.

And yet, at the core, the same. I think travel with the aid of language really helps unlock. You can get fifty five percent off your bubble subscription, a bubble that comes lash x pod, that spelled B A, B, B, E, L, that countless, like pod rules and restrictions apply.

This shows authority by night, sweet and all in one cloud business management system. They manage all the messy stuff that required to run a business, the financial, the human resource, the inventory, if you do that kind of thing. E commerce, all that stuff, all the business related details.

I know how stress I am about everything that's required to run a team, to a run a business that evolves much more than just ideas and designs in engineering, and evolves all the management of human beings, all the complexity of that, the financial, all of IT. And should you should be using the best tools to the job? I sometimes wonder if I have IT in me mentally and skill wise, do will be a part of running a large company.

I think, like with a lot of things in life, as one of those things you should wonder too much about, you should either do or not do. But again, using the best tools for the job is required here. You can start now with a no payment or interest for six months.

Got a nested ET doc slack legs to access their one of a kind financing program that's net sweet 的 com slash legs。 This shows also about to buy inside tracker, a service I used to track biological data that comes from my body to predict, to tell me what I should do with my lifestyle, with my diet, was working and what's not working. It's obvious all the exciting breakthrough that are happening with transformers, with large language models, even with diffusion, all of that is obvious that with raw data, with huge months of raw data, fine tune to the individual, would really revealed to us the signal in all the noise biology.

I feel like that in the horizon, the kinds of leaves in development that we saw, language, and now more and more visual data, I feel like biological data around the corner, unlocking what's there in the multi hierarchical distributed system. That is our biology. What is the telling us? What is the secret holds? What is the thing that is missing that could be aided simple lifestyle changes, simple diet changes, simple changes in all kinds of things that are controllable by individual human being?

I can't wait till that's a possibility in inside trackers taking steps towards that special savings for a limited time when you go to inside track that come slash legs. The show is also brought you by a thea Greens that's now called A G one. IT has the ag one drink.

I drink IT twice a day, at the very least, is an all in one daily drink to support Better health and people performance. I drink a cold is refreshing. It's grounding. IT helps me reconnect with the basics, the nutritional basics that makes this home machine, that is our human body run.

All the crazy mental stuff I do for work, the physical chAllenges, everything, the higher loss of life itself, all of that is somehow made Better, knowing that at least you getting nutrition and jack, at least you're getting enough sleep, at least you're doing the basics, at least you're doing the exercise. Once you get those basics in place, I think you can do some quite difficult things in life. But anyway, beyond all the other is the source of a happiness and the kind of a feeling of home, the feeling that comes from returning to the habit time and time again.

Anyway, it'll give you one month supply a fishel. When you sign up a drink, A G one that house lash legs, this is electing in podcast to support them to please check out our sponsors in the description. And now their friends, his George hearts.

You mentioned something in a stream about the philosophical nature of time. So let's start with a wild question. Do you think time is an illusion?

You know I sell phone calls to come for a thousand dollars and some guy called me and ah like a thousand dollars you talk me for for half and is like a yeah okay so like time doesn't exist and I really wanted to share this with you. I'm like, what do you mean time doesn't exist? right? Like I think time is a useful model, whether IT exists or not, right? Like this quantum physics exist. What doesn't matter, it's about whether it's a useful model to describe reality, is time may be compressed sive.

Do you think there is in a objective reality? Or is everything just useful models? I'm gonna leave at all is an actual thing that we're constructing.

Models for 咋整？

I was hoping you would know.

I don't think that matters.

I mean, this kind of connects to the models will construct a reality with the machine learning, make sure like is is just nice to have useful proximately of the world such that we can do something with that.

So there are things that are real. Colum graph complexity is real. The compressed matter.

讲 sub A T shirt。

And I think hard things are actually hard. I don't think p equals and p oh.

strong words.

Well, I think that's the majority. I do think factory is in p but I don't .

think you the person that folds the majority in all .

walks of life.

So yeah, in the of computer science or one of the sheep, right? But to you times a useful model.

sure.

We talking about on the stream with time. Are you made of time?

I remember half the things I SAT on stream, something someone's going to make a model of all of that.

And it's going to come back to me someday soon. Yeah, probably. Would that be exciting to your sad that there's a George hot model?

I mean, the question is, when the George hot model is Better than George t like I am declining and the model is growing.

what is the metric by which you measure Better or worse in that if you are competing with yourself.

maybe you can just play a game where you have the George hotz answer and the George hot model answer, and ask which people prefer .

people close to you or strangers.

Either one IT will hurt more when it's people close to me, but both will be overtaken by the George od's .

model and be quite painful, right? Loved ones, family members would rather have the model over for thanksgiving .

than you .

or like significant others, would rather sexed. With the with a large language model version of you.

especially when it's fine tune to their preferences.

对， well， that's what we're doing in a relationship, right? We're just fine tuning ourselves but weren't inefficient with the because we're selfish ingredients. So on a language, models can find tunes efficiently, more self lesser.

There's a started word episode where you catch on, jane lost in the delta. Grant makes herself a lover on the holistic, and the lover falls asleep on her army. He snores a little bit.

And you know, jane edits the program to remove that. And then of course, the realization is, wait, this person is terrible. IT is actually all there newlands in corks and slight annoyances that that make this relationship worthwhile. But I don't think we're going to realize that until it's too late.

Well, I think a large language model could incorporate the the flaws in the corks and all that.

still just the perfect amount of corks and law flaws, and make charming without crossing the line.

Yeah, yeah. And that's probably a good mike approximation of the like the percent of time the language model should be a cranky or, I ask, hole a or jealous or all this kind of stuff.

And of course, IT candidate will. But all that difficulty at that point is artificial. There is no more real difficulty. okay?

What's the difference in real artificial?

Artificial difficulty is a difficulty that's like constructed or could be turned off the knob. Real difficulty is like you're in the woods .

and you've got to survive. So if something can not be turned off for the nob, israel.

yeah, I think so. Or I mean, you can't get out of this by smashing the knob with a hammer. I mean, maybe you kind of can you know I into the wild when you know I was under super tramp?

He wants to explore something has never been explored before, but it's the nineties. Everything's been explored. So he's like just not .

going to bring a .

map yeah I mean, no, you're you're not expLoring. You should have got a map to you die those a bridge mile where you were camping.

How does that connect to the metaphor of the knob?

By not bringing the map, you didn't become an explore. You just smashed the thing. Yeah, yeah. The art, the difficulty is still official.

He failed before he started. What if we just don't have existed? Nob.

that maybe is even scarier, right? Like weird exist in the world of nature and nature has been found tuned over billions of years. Um to I have a humans build something and then throw the number away in some grand romantic gesture is harming do you think .

of us humans as individuals? They are like born and die or is that I would just all part of one living orgasm that is earth, that is nature.

I don't think there's a clear line there. I think it's all kinds. Just find the I don't know, I mean, I don't think i'm conscious. I don't think i'm anything. I think i'm just a computer program.

So it's all computation. I think running your head is just as this computer thing running .

in the universe, putting, I think, I believe, the extended church daring thesis.

Yeah but if there seems to be an embodiment to your particular competition, like there's a consistency.

will you a liming models have consistency too? Yeah models that have been orally gift will continually say, you know, like, well, how do I murder ethnic minority so? Well, so I can let you do that. How is a consistency to that behavior?

So chief, like we are chef each other, we we find we're providing feedback and there thereby five tone little pockets of competition. We're still unclear why that pocket of competition stays with you. Like for years it's just kind of felt like you have this consistent set of physical biology um what like whatever you call the neurons firing the electrical system, mechanical signals, all of that that seems to stay there. And IT contains information, stores information, and that information permit through time and stays with you. There's a memory.

It's sticky OK, to be fair, like a lot of the models were building today are very even ri traffic nowhere near as complex as the human .

loss function.

Reinforcement feedback, you know, when I talked about will GPT twelve B A G I. And no, of course not. I mean, cross century losses never gna get you there.

You need, uh, probably oral and fancy environments in order to get something that would be considered like A G I like so I ask like the question about like why I don't know. Like it's just some quality of evolution. I don't think there's anything particularly be special about where I ended up, where humans ended up.

Okay, we have human level intelligence. Would you call the agi whatever we have? G I.

I look, I actually I don't really even like the word A G I um, but general health is defined to be whatever humans have.

Okay, so why can GPT twelve not get us to A G I? So we just like on that.

if your loss function is category cross entropy, your loss function is just try to maximize compression. I ve sound lad， I rap, uh, and I tried to get ChatGPT to help me write raps. And the raps that IT wrote sounded like youtube common raps. You know, you can go on any rapped online and you can see what people put in the comments. And it's the most like mid quality rap you can find is md made is bad, like me, it's every time .

I talk you, I learn .

new words. Yeah.

I like, is is IT like basic so what mean means kind of .

it's like it's like midnight of the curve, right? Ah so is like I like do that intelligence curve um and you have like the dub guy, the mark guy and then the mid guy actually be the mid guy. The worse the mark guys like I put all my money and bitcoin mid guys like you can't put money in bitcoin. It's not real money um .

in all of IT is a genius meme that's another interesting one means the humour, the idea, the absurdity capitated a single image and IT just kind of propagate virally between all of our brains. I didn't get much sleeve last night, so very I sound like come high, but I swear i'm not ah do you think we have ideas? Our ideas have us.

I think that we're going to get super scary memes once the A S actually .

are superhuman. Generate memes. Of course you think you'll make humans laugh.

I think it's worse than that. So um infinite just I have introduced in the first fifty pages is about a tape that you uh once you watch IT once you only ever wanna watch the tape um in fact, you wants to watch the tape so much that someone says, okay, here's a hax off cut off your pink and then I let you watch the tape again and you'll do IT. Ah so we're actually gna build that, I think, but it's not going to be one static tape.

I think the human brain is too complex to be stuck in one static tape like that. If you look at like and brains, maybe they can be stuck a static tape, but we're going to build that using general models. We're going to build the tiktok that you actually can't look away from.

So a tiktok is already pretty close there. But the generation done by humans, the algorithm is just doing their recommendation. But if it's if the argument is also able to do the generation well.

it's a question about how much intelligence is behind IT, right? So the content is being generated by, let's say, one humanity worth intelligence. And you can quantify a human and that's a, you know it's it's X, A flops, yard of flops. We can quantified once that generation is being done by one hundred humanity, you're done.

That's actually scale. That's the problem but also speed yeah and what if it's sort of manipulating the very limited human dopamine engine for porn? Imagine just tiktok. But for porn yeah as I got a brave new world.

I don't even know what they will look like, right? Like, again, you can imagine the behaviors of something smarter than you, super intelligent, in, in, in, in agent, that is dominates your intelligence so much, will be able to completely manipulate.

Is IT possible that IT won't really mini. You just move past. You'll just kind of exist with water exists or the air exists. Yes, see.

and that's the whole eyes safety thing. It's not the machine that's gonna do that. It's other humans using the machine that we're going to do that to you.

Yeah because the machine is not interested hurting humans.

The machine is a machine, yeah, but the human gets the machine. And there's a lot of humans out. They are very interested in manipulating you.

Well, let me bring up at the eza yet kakei, who recently sad where you're sitting, he thinks that A I will almost surely kill everyone. I, do you agree with him or not?

Yes, but maybe for a different reason.

Okay, what? And i'll try to get you to find hope. Or we could find a no to that answer. But why? yes.

okay. Why did nuclear weapons kill everyone?

Question.

I think, is an answer. I think it's actually very hard to deploy nuclear weapons tactically. It's very hard to accomplish tactical objectives.

great. I can note their country when I have a irradiated pile of rubble. I don't want that.

Why not?

Why don't I want in a radiated pile of rubble? Yeah, all the reasons don't want that radiated pilot.

All because you can use that land for for resources. You can populate land.

Yeah, what you want. A, A, A total Victory. A war is not usually radiation and eradication of the people. There is the subjective and domination .

of the people, okay, so you can use this strategically, tactically in war to help you is to help gain a military advantage. It's all complete destruction, right? But the eagles involved is still surprising. So surprising that nobody pressed a big red button.

It's somewhat surprising. But you see, it's a little red button that's gonna pressed with a eye that's going to, you know and that's why we die. It's it's not because the A I IT is anything in the nature of A I, it's just the nation, humanity.

What's the algorithm? Hind a little red button. Like what possible ideas do you have for the how human species ends?

sure. So I think the most 啊 obvious way to me as war adding， we end up amusing ourselves to death, we end up all staring at that infinite tiktok and forgetting to eat. Maybe maybe it's even more benie than this. Maybe we all just stop reproducing. Now, to be fair, it's probably aren't hard to get all of humanity.

Yeah, yeah. IT always go like the the interesting thing about humanity is a diversity organisms in general. This a lot weird dos out there. Two of them.

We're sitting here. I mean diversity in humanity as we do respect. I wish I was more weird, not like i'm kind of a look and drink smart waterman that's like a coal product, right?

Do you want corporate?

Just a no. The amount of diversity in humanity, and I think is decreasing, just like all .

the other biodiversity on the planet.

Donald, in china.

yeah.

yeah. No way is the interconnected us. That's that's that's doing that.

Are this interesting? So everybody starts relying on the connectivity of the internet. And over time, that reduces the diversity, the intellectual diversity. And then that gets you everybody, into a fund that still going to be a guy in texas.

There is. And yeah, that, to be fair, do I think A I kills us all, uh, I think A I kills everything we call that society today. I do not think that actually kills the human species. I think that's actually incredibly hard to do.

Yeah a society like if you start over that tRicky most of us don't know how to do most things yeah.

but some of us do and they'll be okay and they'll rebuild after the great a eye.

what's rebuilding looks like like how much do we lose that? What is human civilization done? That's interesting in computing engine electricity. So power and energy, that's interesting. Like how to harness energy.

They going to be religiously against that.

Are they going to get back to a like fire?

sure. I mean, not be a will be little bit like, you know, some kind of amah looking kind of thing. I think I think they're going to have very strong taboos against technology.

嗯， like technologies is the second new religion. Technologies is the devil, and nature is god. Sure to close nature, because you really get away from my eye. If IT destroyed, ninety nine percent of the human species isn't somehow have a hole like a stronghold.

What what's interesting about everything we build, I think going to build super intelligence before we build any sort of robust in the A I. We cannot build an A I that is capable of going out into nature and surviving like a um like a bird. And Albert is an incredibly robust organism. We've built nothing like this. We haven't built a machine that's capable of free producer.

yes, but there is a in our work like robots a lot. Now I have a bunch of them. There are mobile.

They can't .

reproduce. But all they need is I guess you're saying they can repair themselves, but you have a large number if you have like a hundred million of them.

let's just focus on them reproducing, right? Do micro chips in them, okay. Then do they include fab?

no.

Then how are they going to reproduce?

The other IT doesn't have to be all on board, right? They can go to A A factory to a repair shop.

Yeah, but then you're really moving away from your business. Yes, all of life is capable of reproducing without needing to go to a repair shop. Life will continue to reproduce in the complete absence of civilization, robots or not. So when the if, if the A, I, A park lapse happens, the eyes are gonna probably die out, because I think we're gonna, again, super intelligence long before we get our business.

What about if you just improve the fab to wear you? You just have a three d printer that can always help you.

Will I be very interesting? I'm interested in building that.

Of course, you are. You think how difficult that problem to have a robot that basically can build itself very.

very hard. I think .

you've mention this like to me, or somewhere where people think is easy conceptual.

And then they remember that you're going to have to have .

a fab yeah on board. So three d printer that prints a three d printer? Yes yeah on legs yeah hard.

Well, because I mean, a three d printer is a very simple machine, right? Okay, you're going to print chips. You gona have an atomic printer.

Are you gonna be the silicon?

Are you gonna dge the silicon?

You're gonna to have a very interesting kind of fab. If you want to have a lot of computation on board. They can do .

like structural .

type of robots. They're dumb.

Yeah, but actual type of robots aren't going to have the intelligence required to survive in any complex environment.

What about like ants type systems? We have like trillions of them.

I don't think this works. I mean, again, like ant at their very core are made above cells that are capable of individually reproducing .

that do quite a lot, a lot of competition that we're taking for granted.

It's not even just the computation is that reproduction is so inherent OK. Like there's two stacks of life in the world. There's biological stack.

The biological stack starts with reproduction. Reproduction is that the absolute core, the first proto ornate organisms were capable of reproducing. The silicon on stack, despite as far as its come, is nowhere near being able to reproduce.

Yeah so the the fed movement, digital fabrication, fabrication in the full range of what that means is still in the early stages. Yeah, you're interested in this world even .

if you did put a fab on the machine, right? Let's say OK, we can build up to you humanity. We can probably put all the press that build all the machines and the ABS also in the machine.

So first of, these machines can be absolutely massive. I mean, we almost have, like, think of the size of the thing, require to reproduce a machine today, right? Like, is our civilization capable reproduction? Can we reproduce our civilization? Mars.

if were, to construct a machine that is made up of humans like a company, they can reproduce itself. Yeah, I don't know. He feels like like, like one hundred fifteen people.

I get so much hard than that.

twenty.

I believe that twitter can be run by fifty people 啊啊 i think that this is gonna take most of like it's just most of society， right? Like we live in one globalized world.

no. But you not interested in running where you're interested in seeding, like you want to see this civilization and then humans can like.

oh, OK, you're talking about, yeah okay. So you're talking about the humans reproducing and like basically like what's the smallest staying colony of humans? Yeah, yeah. Okay, I fine. But you're not going to be making five hundred of chips over time.

They will. I think you're being I like we have to expand our conception of time here. Come back to the regal at time.

Scale, mean, over, across. Maybe hundred generations were back to make a chips. No, if you seed the colony correctly.

maybe, or maybe they will, what our colony die out over here and be like when not making chips. Don't make chips, not be.

you have to see that colony correctly.

Whatever you do, don't make chips. Chips were LED .

to their downfall. What that is the thing that humans do, that they come up the constructed devil, a good thing in a bad thing, and they really stick by that. And they murder each other over that, want to ask on the room .

who is everybody.

and usually makes tattoos and nice branding.

Need that ask, ask a question, right? You really works really hard today to get rid of that as well. But I think they might be important.

Yeah, that's whole freedom of speech thing. It's the freedom of being an asso seems kind of important, right, man. This thing, this fab, this human fab that we constructed, this human civilians, pretty interesting.

Now it's building artificial copies of itself, or artificial copies of various aspects of itself. That seems interesting, like intelligence. And I wonder what that goes I like .

to think is just like another stack for life, like we have, like the biosphere life, like a BIOS stack life and silicon stack life.

But IT seems like the ceiling, but there might not be a ceiling in, or at least to see as much higher for the for the silicon stack.

Ohh no I don't I we don't know what the ceiling is for the bio stack either. The bio stack just seem seem to move slower. Um you have a mores law, uh which is not dead despite many programs tions are in .

to o stack the in the silk .

and don't have anything like this in the biosphere. So I have A I mean that I post I tried to make a me didn't work too well but I posted a picture of you know an old reagan and job and you look, this is one thousand nine and eighty, and this is twenty twenty year and these two humans are basically like the same, right? There's no, there's no like like there.

There's been no change in humans in the last forty years. Yeah and then I posted a computer from one thousand and eighty and a computer from twenty twenty. wow.

Yeah, with their early, early stages, right? Which is why you said, we said the fab, the size of the fab required to make another fab is like very larger right now. But computers were very large eighty years ago, and they got pretty tiny and people starting to want to wear them on their face.

In order to escape reality, that's a thing. In order to be live inside the computer, put the screen right hair. I don't have to see the rest of the U. S. S.

I've been ready for a long time.

You like virtually love. Do you want to live there? Yeah, yeah, part of me does. To how far away are we? They think.

judging from what you can buy today, far, very far.

I gotto tell you that i've had the experience of mats coltec avatar. Where is an ultra high resolution scheme? IT looked real. I mean, the headsets .

just are not quite like a resolution yet. I haven't put on any head that where i'm like, oh, I this could be the real world whether when I put good headphones on, audio is there I like we can reproduce audio that I like. I'm actually in the jungle right now. I I close my, yes, I can tell him not yeah.

But then is also smelling all that kind of stuff. sure. I don't know. I the the power of imagination, the power of the the mechanism in the human mind that fills the gaps. The kind of Richardson wants to make the thing you see in the virtual world real to you. I believe in that power our .

humans want to believe.

yeah. And what? What if you're lonely? What if you're sad? What if you're really struggling in life? And here's a world where you don't have to struggle anymore.

Humans want to to believe so much that people think the large language models are conscious. That's how much humans wanting .

believe strong words he's throwing left and write hooks. Why do you think large language models .

are not conscience? I don't think concious.

Oh, so what is consciousness then?

George house, it's like what IT seems to mean to people. It's just like a word that I P S used for souls.

Sure, but that doesn't mean soul is in word. If consciousness is .

a spectrum, i'm definitely way more conscious than the large language models. I think the large language handles are less contrast than a chicken.

One is less things seen .

a chicken ah in my amy like come mosk. No.

like a living check.

Living check is walking around. My amy is crazy.

like on the street. Yeah like a chicken, a chicken. Yes, right? right? It's trying to call you all like a good journalist and I guess shut down, okay? But you don't think much about this kind of subjective feeling that IT IT feels like something to exist. And as an observer, you can have a sensitive entity. It's not only intelligent, but as a kind of subjective experience of its reality, like a self awareness that is capable of like suffering, of hurting, of being excited by the environment in a way that's not merely kind of an artificial response, but a deeply felt one.

Humans want na believe so much that if I took a rock and and a sharply into a sad face on the rock they think the rock is said.

yeah and you're saying when we look in the mirror we we apply the same smile .

facebook .

rock is not weird though that you not conscious .

so but you do believe in .

consciousness. It's just it's unclear okay. So you do it's like a little like a symptom of the bigger thing that's not that important.

Yes, it's interesting that like the human system seem to claim that they are contrast. And I guess I kind of like, says something in a strait up like, okay, what do people mean when, even if you don't believe in consciousness, what people mean when they say consciousness finally means to IT?

What's your favorite .

to IT pizza .

chess piece worth the toppings? I, apple.

any help on feel bad.

What's the best? What's the bus pizza? What we talked about like you like a crab pizza.

chicago deep dish cheese .

into a deep dish 的 批 and IT feels like you were starving。 You have an eat for twenty four. You just bite in and you're hanging out .

with somebody that matches .

IT feels like something i'm George, mother fucking hearts eating a fucking chicago deep dish pizza. There's just a full peak light living experience of being human, the top of the human condition. IT feels like something to experience that why does he feel like something that's consciousness isn't 呢？

If that's the word you want to use to describe IT, sure. I'm not going to deny that feeling exists. I'm not going to deny that I experience that feeling when I guess what I kind of take issue to is that there is some like like how does that feel to be a web server to follow force? hurt.

But yeah.

how would you know what suffering looked like? Sure, you can recognize the suffering dog. Because where the same stack as the dog, all the bio stacks of kind of especially mammals, you know it's really easy. You can game recognizes .

game .

yeah versus the silicon tax stuff. It's like you have no idea. You have used IT. Well, the little thing has learned to mimic, you know but then I realized that that's all we are too. All of the little thing has learned, dominic.

Yeah I guess you four, four, four could be could be suffering but it's so far from our kind of living orgasm, our kind of stack. But IT feels like A I can start maybe making the biological stack but but but because is trained, trained IT. yeah. And so in that maybe that's the definition of consciousness, is the BIOS conscious. The definition .

of conscious ness is how close something looks to human. Sure, i'll give you down.

No, how close something is to the human experience.

sure, is a very, very enticement c definition. But sure no, I don't mean to like I think there's a lot of value in IT. Look, I just started my second company. My third company will be A I girlfriends. I mean.

I want to find out what your fourth company is because I think once you have A I girlfriend, it's a boy doesn't get interesting. But maybe let's go there.

I mean, the relationships with the eyes that's creating human like organisms, right? And part of being human is being conscious, is being having the capacity to suffer, having the capacity to experience this in such a way that you can empathy the air, is can empathy with you and you can empathy with IT, or you can project your anthropy c sense of what the other entity is experiencing. And an A I model would need to um yeah to create that experience inside your mind. And IT doesn't seem like difficult yeah but okay.

so here's where IT actually gets totally different, right? When you interact with another human, you can make some assumptions.

Yeah when you interact .

with these models, you can you can make some assumptions that that other human experiences, suffering and pleasure in a pretty similar way to you. Do the golden rule applies with a nyi model? This isn't really true, right? These these large language models are good at fooling people because they were trained on A A whole bunch of human data and told a memc.

But if if the air system says, hi, my name is Samantha has a bad story. Yeah, want to college you here and there. Maybe you want to create this in the eye system.

I made some chap out. I give him back. Story was a lot of fun. Are so happy when .

love came yeah we will talk about them, will talk about all that. But like, you know, the rock with a smile face yeah well, this this seems pretty natural for you for you to antipa fy that thing and then start dating IT and before you know, if you're married and have kids.

rock with a rock.

this pictures on instagram with you in iraq and face.

to be fair, like, you know, something that people generally look for. The look for someone today is intelligence and some form in iraq doesn't really have intellect, only a pretty desperate son would take iraq.

I think. World desperate, deep down.

Oh, not rock level desperate.

Not rock level desperate. But A I level desperate. I don't know. I think all of us have a deep loney. Ss IT just feels like the language models are there.

Oh, I agree. And you know what, I won't even say this. So cnc ally, I will actually say this in a way that, like, I want my friends, I do yes, but I would love to you know, again, I the language one is now is still little like people are impressed with these GPT things.

And I look like or or like or the the copilot, the coding one, and i'm like, okay, this is like junior engineer level and these people are like five or level artists. And copywriters like, okay, greate, we're got like fiver. And like junior engineers, okay, cool. And this is just the start and he will get Better, right? Like I would I can't wait to have A I friends who are more intelligent than I am.

So five years, just a temporary is not the ceiling?

No, definitely not.

Is IT is account is cheating when you're talking to an eye model, emotional cheating.

That's that's up to you when your human partner .

to define I have .

you ah have have that conversation I guess .

alright I mean integrate that was a with porn and .

all this no I the point yeah why I think people in relationships have different views .

on that yeah but most people don't have like serious open conversations about all the different aspects of what's cool and was not. And if he s like A I is a really weird versac to have.

one is a good branching off like these things. You know, one of my scenario that I put in my chatbot is, you know a my score name, lexi. She's twenty.

SHE just moved out to L A. SHE wanted to be an actress, but SHE started doing only fans instead. And you're on a day with her. enjoy.

Man, yeah. And so was that if you're actually dating somebody in real life, is that cheating? I feel like that gets little weird.

I guess real weird. Like, what are you a lot to say? And A I bought imagine having that conversation was a significant I mean.

these are all things for people to define in their relationships. What IT means to be human is just going to start to get weird.

especially online. My, how do you know like it'll be moments when you have what you think is a really human interactive with on twitter for years, and you realized, is not.

I spread. I love this. Mean, uh, heaven banning, you know what? Shadow banning, yeah, shout of banning. Okay, you post. No one can see IT having banning, you post, no one can see that. But a whole lot of a eyes are spot up to indirect with you.

Well, maybe that's what the way human civilization, and is all of a heaven banned.

There's a great called my little pony. Friendship is optimal. It's a safer story that explores this idea.

Friendship is, friendship is to mal. Yeah, i'd like to have some at least of intellectual realm, some A I friends that are good me. But the the romantic ground is weird, definitely weird, but not a realm of the the kind of weird that the human suiza capable. I think.

I think I want IT. Look, I want IT if no one else wants IT, I want IT yeah .

think a lot of people probably want IT. There's a deep low .

Venus and i'll fill their loneliness and you know, just will only advertise to you some of the .

time yeah maybe the conceptions of monogamy y change to like I grew up in a time like I value monogamy but maybe that's a silly notion when you have arbitrary .

number of A I systems. This is this interesting path from rationality to polyamory that doesn't make answer me for you.

but you're just a biological organism who was born before I E, the internet really took off.

The crazy thing is, like, culture is whatever we define IT as right? These things are not, you think like is a problem in moral, physical. Hy, right? There's no like, like, okay, what is might be that like computers are capable mimicking, uh, you know, girlfriends perfectly.

They passed the girlfriend turning test, right? But that doesn't say anything about what that does not say anything about how we ought to respond the realization. That doesn't say we ought to get rid of anarchy. That's a completely separate question, really religious one girlfriend .

touring test, I want what that looks like. Are you writing that? Um will you be the the the Allen touring of the twenty four century that writes the other golf .

in tortoise SE my AI girlfriend? Their goal is to pass the girlfriend turn test.

Not but there should be like a paper that kind of defines the test or so. I mean, the question is if it's deeply personalized, there's a common thing that really gets everybody .

yeah I mean, you look at our company, we don't like everybody. We just have to get large enough client.

And what are you already already thinking? Company, let's before we go to company number three and company number four, let's go to company number two, right? Tiny corp, possibly one of the greatest names of all time for company I.

You've launched a new company called tiny corp, at least the development of tiny grad. What's the origin story of tiny corp. And tiny grad?

I started tiny grad as, like, a toy project just to teach myself. Okay, like what is a convolution? Uh, what are all these options you can pass on on what is the devil? Convolution right? Very simple carpathia microgram very more. And then I started realizing, I started thinking about, like AI chips. I think about chips that run A I and I was like, well, okay, this is going to bear a really big problem if in video becomes a monopoly here um how long before in .

video is nationalized so you are one of the reasons to start tiny corp. Is to chAllenge in video.

It's not too much to chAllenge in video. I I actually I like and video and it's to make sure power stays desensitized .

yeah and here's uh, computational power until you n video kind of locking down the competition power the world.

if NVIDIA becomes just like tanks s Better than everything else. You're giving a big advantage to somebody who can secure envy as a resource. You in fact, if Jenny watches this podcast, he may anna consider this. He may anna consider making sure his companies is not nationalized.

Do you think that's an actual threat?

Oh, yes.

No, but there's so much. Uh.

you know there's M D, so we have a video and M D great.

right? But you you know think there's like a push towards like selling like google selling keep use as something like this. You don't think there's a push for that.

Have you seen IT google loves to rent U. T, P.

S. IT doesn't. You can buy IT, a best buy.

So I start work on a on chip was like, okay, what's IT going to take to make a chip? And my first notions were all completely wrong about why, about, like how you can improve on G. P.

S. Uh, and I will take this. This is from a jim keller on your forecast. And this is one of my absolute favorite descriptions of computation.

So there's three kinds of computation paratime that are common in the world today are there CPU and CPU can do everything. CPU can do ad and multiply. They can do loading store and they can do comparing branch.

And when I say they can do these things, they can do them all fast, right? So comparing branch er unique to CPU and what I mean by they can do them fast as they can do things like branch production and speculative execution and spend tons of transistors and it's like super deep reorder buffers. In order to make these things fast, then you have a simple of computation model.

G P S. G P S can't really do compared branch, I mean, they can, but a horrendously slow. But G P S can do arbitrary loading store.

But G P S can do things like xd reference. why? So they can fetch from arbitrary peces memory. They can fect from memory that is defined by the content of the data. On the third model of competition.

D, S, and D, S, P, S are just add a multiple like they can do, loads stores, but only static loads stores, only loads and stores that are known before the programme runs. And you look at neural networks today, and ninety five percent of neural networks are all the D, S. P.

They are just statically scheduled ads and multiples. So tiny god really took this idea and and i'm still working on IT to extend this as far as possible on every stage of the stack has turned combines and python has turning complete us. And then we take path on, we go to class, which turn to and maybe plays calls into some internals, which are turning complete internals. Go through L, V, M, which to comte into P, X, which to which the process on to get turn completely out of the stack entirely. Because once you get rid of turning completely, you can reason about things rises serum in the holding problem, do not apply to have more machines.

Okay, what's the power in the value of getting turn completely out of out of? Are we talking about the hardware or or the software?

Every layer of the stack every day, every layer of the stack moving turn completely ness allows you to reason about things, right? So the reason you need to do branch prediction in the C. P. U. And the reason it's prediction and the branch predictions are, I think they're like ninety nine percent on CPU.

Why do they get one percent of them wrong? Will they get one percent wrong? Because you can know right that's the halting problem is equivalent to the halting problem to say whether a branch is going to be taken or not um I can show that but the ad mo machine, the neural network runs the identical compute every time the only thing that changes the data.

So when you realize this, you think about, okay, how can we build a computer and how can we build a stack that takes maximum advantage of this idea? Uh, so what makes tiny grad different from other neural network libraries is IT does not have a prevent of Operator even for metric notification. This is every single one. They even have privative Operations.

things like convolutions.

So no much, no. Well, here's what to met my less. So I use my hands to talk here. So if you think about a cube and I put my two major aces that are multiplying on two faces of the cube, right, you can think about the matrix multiple as, okay, the n cubed. I'm going to a multiple for each one of the cubed. And then I to do with some, which is reduce up to here to the third face of the cube and match your multiply matrix. So what a matrix multiply is is a bunch shape Operations, write a bunch of premise, three shapes, and expands on the two materials are multiply and cubed, a reduce and cubed, which gives you great matrix.

Okay, so what what is the minimum number of Operations that can accomplish that if you don't have that mall as a primitive?

So tiny grad has about twenty um and you can compare tiny grads are upset or I R two things like x ela or prim torch. So excEllent prim torch our ideas we're like, okay torch has like two thousand different kernels um pitch two point o introduced prim towards for chanell to and fifty uh tiny grad order language to twenty five it's it's ten x less than x allier from torch and you can think about IT as kind of like a risk for a sister, right? These other things are sister like systems, a ten to risk and .

risk one.

Risk architecture gonna change everyday. Nineteen ninety five hackers.

really, that's the natural thing.

Angela joe delivers the line. Risk architecture is gona change everything in nineteen ninety five. And here we are with arms in the phones and everywhere.

Wow, I love when movies actually have real things in them, right? Okay, interesting as this is like you think of this is the risk architecture of, i'm .

stack twenty five.

What what can you can you go through the ah the four other types?

Sure um okay so you have your ops which take in uh a tensor and return a tensor of the same size and do some unity up to x log a recipe al sign right they taken one in their points SE really yeah .

review um almost activation .

functions are unity ops um some combinations of unity ops together is still lunary ops um then you have vary ops.

Binary ops are like uh point one edition multiple ation division compare uh IT takes in two tensor of equal size and outputs one tenser um then you have reduce ops, reduce up will like take a three dimensional cancer turn in into a two dimensional tenser or three dimensional cancer turn into is zero dimensions tenser think like a sum or max are really the common ones there and then the four type is movement tops. And movement tops are different from the other types because they don't actually require computation. They require different ways to look at memory. So that includes reshapes, promotes, expands, flips, doesn't the main ones .

probably so that you have enough to make a matter .

more and convolution and every convolution you can imagine dillaway convolutions started convolutions transported .

convolutions you're write on, get hub about laziness, showing a map more mature motivation. See how, despite the style, IT is used into one kernel with the power of laziness. new. Elaborate on this power.

laziness? sure. So if you type in PyTorch a times b plus, uh, what this is going to do is is going to first multiple A N B, A N B and store that result in memory. And then that is going to add c by reading that result for memory. Reading see for memory uh writing that out to memory um there is way more loads and stores to memory than you need there. If you don't actually do eight times as b as soon as you see IT, if you wait until the user actually realizes that tenser until the lazy is actually resolved um you confuse that plus this is like it's the same.

my house call works. So um what's the process supporting a model into degrade?

So ti grads front end looks a very similar PyTorch. I probably could make a perfect are pretty close to perfect interrup liar if I really wanted to. I think that there's some things that are nice are about tiny grads and tax and PyTorch, but the front looks very torch like are you can also loading on its models. Um we have more onest test passing than core mail.

okay.

so will pass on exchange types.

What what about like the developer experience of tiny grad, what he feels like, what a versus PyTorch. By the way, I really like pitch.

ch. I I think that is actually a very good piece of software. Um I think that they've made a few different trade ffs. And these different trade ffs are, well, no turning grab, take different bath. One of the biggest differences is is really easy to see the cornels that are actually incentive G P U, right? If you run pyto ch on the G P U, you like do some Operation and you don't know what kernel l and you don't know how many kernels run, you don't know how many flops were used, you don't know how much memory access were used. Tony, grade type debug equals two and I will show you in this beautiful style um every cornal that's run how many flops and how many bite.

So can you just linker on what problem tiny grad solves? Tiny grad .

solves the problem of porting new and accelerators quickly. One of the reasons are tons of these companies. Now I think i'm sqa marked graph core to zero, right sabas ten stories are rock all of these m acela atr companies.

They build chips. The chips were good. The software was terrible. Uh, and part of the reason is because I think the same problem problem happening with the dojo, it's really, really hard to write a, because you have to write two hundred fifty kernels and you have to tune them offer performance.

What is jim jm, color? Think about thy grand. You guys hung out quite a bit.

So he, you know, he was involved. He's involved. And what's his praise and what is criticism, what you're doing with your life?

Look, my prediction for tense torrents is that there gona pivot to making risk. Five tips, c pu.

c pu. Why A I accelerators .

are software problem? Not really harder problem.

always. Thanks for your own dick. You think that diversity of A I accelerators in the hardware space is not going to be a thing that exists long term?

I think what's gonna happen is if I can fin IT, okay, if you're trying to make an eye accelleration or you Better have the capability of writing a torch level performance stack on G P S. If you can write a torch stack on n via G P S. And I mean, all the way, I mean, down to the driver, there's no way you going to be to write IT on your chap because your chips worse than novita GPU. The first version, the chip tape .

out is definitely worse. You that's tough.

yes. And not only that, actually the chip that you tape out almost always because you trying to get advantage over in video, you're specializing the harbor more. It's always harder.

The right software r for more specialize hardware like A G P S, pretty genre. And if you can't write an a video stack, there's no way you can write a stack for your trip. So my proof, tiny grad is first, write a performance and video stack, or we're argee empty.

So you did say a few to end video a little bit would love .

yeah so ah you know .

what ah your mass fan, a risk a risk fan, a mass fan. What's the hope that M D has? You did build with the M D recently that I saw. Uh, how does the, uh the seventy nine hundred X T X compared to the RTX8？ Let's start with the .

fact that the hundred x nal drivers don't work. And if you run demo apps and loops IT panic the kernel.

okay. So this is a softer issue.

lisa, who responded to my email, uh, I reached out. I was like, this is, you know, really like, I understand if you are seven by seven transport SE winner grad, calm is slower than in videos, but literally when I run demo apps in a loop, the corner panic.

So just adding that loop.

yes, I just I just literally took their democrats and wrote, like wild ue semi collin do the APP semi collen done in a bunch of screens. This is like like the most primitive fuz testing.

Why do you think that is they're just not seeing a market in the in the machine learning. They're changing.

They're trying to change. They're trying to change. And I had a pretty positive interaction within this week, last week.

I they want you like I give up on A M D. Like this is that their driver doesn't like i'm not going i'm all go with intel G P S. Intel G P S have Better drives.

So you kind of spearheading the diversification of G P S.

yeah. And i'd like to extend that diversion ation to everything. I'd like to diversify the right. The more my central thesis about the world is those things that centralize power and y're bad, and there's things that decentralize power and they are good. Everything I can do to help decentralize power, i'd like .

to do so. You really worried about the centralization. And video is interesting and you don't have a fundamental hole for the the proliferation of asics. I except in the cloud.

i'd like to help them with software. Now actually, there's only the only asset that is remotely successful is google TPU. And the only reason that successful is because a google wrote a machine learning for me work. I I think that you have to write a competitive machine learning ramework in order to be .

able to build asic. You think meta with PyTorch builds competitor.

I hope so. Have one. They have an internal one internal.

I mean a public facing with a nice cloud interfaces on, I don't want a cloud. You don't like cloud. I don't like cloud. What do you think is a fundamental .

limitation of cloud, fundamental limit .

of people? And you don't like the man to have all the power.

right?

And right now, the only way to do this with the D G P, S, if you want performance instability, interesting, which to cost the investment emotionally, to go with the ams. Well, let me get sort of an a change. Ask you, you've built quite a few PC. What's your advice on how to build a good custom PC for, let's say, for the different applications they use for gaming, for a machine learning?

You shouldn't d, you should buy a box from the tiny corp.

I heard rumors, whispers about this box in the tiny corp. Is what this thing look like. What what is that? What is that called?

I got the tiny box, tiny box, fifteen thousand dollars. And it's almost a paid a flop of compute. It's over a hundred gigabits of GPU RAM. It's over five terabytes per second of GPU memory band with when I put like four n vm is in in in RAID, you're going to get like twenty thirty minutes per second of drivers band weth. I'm i'm going to build like the best deep learning box that I can that plugs into one wall outlet.

okay. Can you go those picks again a little bit from from memory yeah so .

it's almost to pay the top of computer.

I so M D and l today .

i'm leaning to M D um but we're pretty agnostic to the type of computer. The the main limiting spec is a one hundred twenty volt fifteen eight circuit IT. okay. I mean that because in order to like like there's a blog over there, right, you have to be able to plug IT in. Um we're also gona sell the tiny iraq, which like what's the most power you can get into your house without arousing suspicion uh and one of the one of the answer is an electricity charter.

Where where does the red go?

You go out .

and sting the car charger.

A wall out with is about fifty hundred watts. A car charger is about ten thousand.

What is the most demand of power you can get your hands on without a rousing suspicions, right? George hot, okay. Um so the the tiny boxing he said and be amused and RAID um I forget what you said about memory, all that kind of stuff.

okay. So I what about what? G, P, S.

again, probable probably seventy nine hundred X T access, but maybe thirty nineties, maybe seven seventy because .

you're flexible or still expLoring.

I'm still expLoring. I I want to deliver a really good experience to people. And yeah, what G, P, U, I end up going with again, i'm leaning to M D IT.

We will see, you know, in my email what I, what I said, M, D is like just dumping the code on github is not open source. Open source is a culture. Open source means that your issues are not all one year old stale issues.

Open source means developing in public. And if you guys can commit to that, I see a real future for A M, D. As a competitive video.

Well, i'd love to get a tiny box. M, I, T. So whenever it's ready.

let's do IT. I for me on my hundred fully refundable promoters is IT going .

to be like the cyber truck.

Is onna take a few years or now i'll try to do a fast? It's a lot of employment to truck.

Well, there is complexity is not to just the put the thing together, but like shipping in all this kind of stuff.

And the thing that I want to delivered the people out of the box is being able to run sixty five billion parameters ma in fp sixteen in real time. And like good, like ten token s per second or five token .

s per second or something just IT works was running uh or something like lama experience or I think .

fcon is is is the new one experience a chat with the largest language model that you can have in your house?

Yeah from from a while plug.

from a while plug. Actually, for inference, it's not like even more power would help you get more.

I think even we can get more.

This just the bigger the biggest model released is, is sixty five billion primary along as first, I know.

So IT sounds like tiny box will naturally pivot our company number three, because you could just get the girlfriend and um or boyfriend that was .

that the boyfriend is harder.

I think 这是 very biased statement and think a lot of people to see what what why is IT harder to replace a boyfriend and a the girlfriend with the artificial ell？ M, because women are attracted .

to status and power, and men are attracted to youth and beauty. No, I mean this what I mean.

but what? both? There could be a mimicry. Easy to the language model.

No, no. Machines do not have any status or real power.

I don't know. I think you both well, first, like you're using language mostly to to communicate youth and beauty and power and that but status.

fundamental isms are some game, right?

Where is you can beauty or not? No, I think status is a narrative you can construct. I I think status is real.

I don't know. I just think that that's why a tarter, you know yeah.

maybe IT is my biases. I thinks this is way easier to fake.

I also think that you know, men are probably more desperate and more likely to buy my products. So maybe there a Better target market.

Desperation is interesting. Easier food. Yes, I I can see that. Yeah.

look and look and you can look at point viewing numbers, right? A lot more men watch point than women.

Yeah, that's quite at this. Well, there's a lot of questions and answers you can get there anyway with with the tiny box. How many G, P, S in tiny box sex? Man, i'm i'll tell .

you why I sex yeah .

ah .

so aimed epic processors have hundred twenty eight lanes of P C. Um I want to leave enough lanes for some uh, drives and I want to leave enough lanes for some networking.

How do you cooling for something like this?

That's one of the chAllenges. Not only do I want the cooling to be good, I wanted to be quiet. I want the tiny box to people to sit comfortable in your room.

This is really going to ours. The girlfriend thing. So because because you want to run the L M.

i'll give, i'll give a more. I mean, I can talk about how relates to company. Number one.

come on A I, well, we are oh, quite because you may be potential car.

No, no, quite because you wanted put this thing in your house, and you wanted to go exist with your fit streaming at sixty. D, B, you don't want that your house.

you will kick IT out. D, to be here.

I want like forty.

forty five. So how do you make the cooling a quiet? That's an interesting problem itself.

a key tricks to actually make a big ironically, it's called the tiny box yeah but if I can make a big and a lot of that noise is generated because, uh, of high pressure are if you look like a one you server, a one you server has the super high pressure fans that like super deep and a the genus versus if you have something that's big, what I can use a big and you you know, they call big s fans the ones that are like huge on the ceiling .

and they're completely silent. So tiny box will be big.

IT is the, uh, I do not want you to be large according to U. P. S. I want you to be shipped as a Normal package, but that I can strengthen.

interesting. Well, the the fans stuff can can not be assembled on location or no, no, no, I should be. Well, here.

I want to give you a great out of the box experience. I want you to live this thing out. I wanted to be like, like the mac, you know, tiny box.

the apple experience. Yeah, I love IT. okay. And a tiny box to run tiny grad like what what do you vision this whole thing to look like or talking about like Linda s, with the full soft engineering environment and just not photos for tiny grad?

Yeah, we did a poll. If people want you, bound to our arch were going to stick with you. Bond.

oh, interesting. What's your favorite? The flavor of likes. Want to, I like you. Bound to my day.

How will we been on the meat? So how do you you gotten lama into tiny grad? You ve got a stable, the fusion.

Tiny grad, what was that like you comment on? Like what um what are these models? What's interesting about voting them? What's we I look what the chAllenges, what what's naturally is easy. There is a really simple .

way to get these models into ti grad and you can just export them as honey and then time you got can run on ics. Um so the ports that I did of lama stable fusion and now whisper are more academic to teach me about the models, but they are cleaner than the python versions. You can read the code.

I think the code is easier to read, its less lines. This is a few things about the way time you got rights next years. Here's a complaint I have about pyrot's and and not rely was a class, right? So when you create a module, you'll put your N N values as in a net.

And this makes no sense. Rally was completely stateless. Why should that be a class?

But that's more like a soft engineering thing. Or do you think he has a cost? Some performance .

oh a some performance. Um but yeah no, I I think that is it's that's what I mean about like tiny grad front and being cleaner.

I see waiting about moto. I know you're been paying attention the program language that does some interesting ideas that kind intersect uh.

tiny grad. I think that there is a spectrum and like on one side you have mojo on the other side of like G G M L um G G M L. Is this like we're onna run lama fast on mac? Okay, we going to expand out to a little bit, but we to like debt first right? Mojo was like we're gona go breakfast.

We're going to go. So why that we're going to make all of piton fast and tiny grades in the middle time? Again, we are going to make neural .

networks fast. Yeah but they they try to really get IT to be fast. Compiled down the specifics, uh, hardware and make that compilation step as flexible and resilient as possible.

If they've turned completely .

and that will miss you, that's where you is somewhere in the middle. So you are actually going to be targeting some accelerated some like some, some number, not one. My goal is.

step one, build an equally performance stack of PyTorch on in video and AMD, but with way less lines. And then step two is okay, how we make an accelleration or right? But you need step one. You have to first build the framework before you can build the accelerator.

Can you explain M A perf. Was your approach in general to benchmarking tiny grad performance.

So i'm much more of I like build IT the right way and worry about performance later. Um there's a bunch of things where I haven't even like revise ave into performance. The only place where tiny gradus competitive performance was wise right now is on cocom G P S. So ti grade I actually use an open pilot turn the model. So the driving model is is, is tiny grad.

When did that happen? That transition .

eight months go now and it's too exact and welcomes library.

What's the hardware of open ah that open pilot s on the of the company .

is a small dragon forty five OK. So this is using the G P U. So the G P U in the geno G P U, there's like different things.

There's a really good microsoft paper that talks about like mobile G P U and why they are different from descript G P S. Um one of the big things is in the desk cup GPU. You can use buffers uh on a mobile G P U image textures or a lot faster.

and a mobile G, P, such is a limit. okay. And see, you want to be able to leverage that.

I want to be able to leverage IT in a way that it's completely generic, right? So there's a lot of this, show me, has a pretty good open source library. Mog PS called me where they can generate where they have these kernels, but they are all encoded, right? So that's great if you're doing three by three calves. That's great if you're doing dance map mouse. But the minute you go off the beaten in path at tiny bit, well, your performance is nothing.

The invention open pilot, look to get an update in the company. Number one, come the eye. How are things going there in the development of a simula was driving.

You know, almost no one talks about F, S, D anymore, and even less people talk about open pilot. We have the problem like we solved IT years ago.

What's the problem exactly? Well, what does solving IT mean? Solving means how .

do you build a model that, uh, output to human policy for driving? How do you build a model that, given a reasonable set of sensor, is output to human policy for driving? Uh, so you have, you know, companies that women cruse with your hand coding, these things that are like quality human policies.

Then you have tesla and maybe even even to more of an extent, comma asking, okay, how to do just lean business data. The big thing that we're doing now, and we just put IT out on twitter at the beginning of coma, we polish the paper called learning a driving simulator. And when everything worked, was IT a IT, was a lot encounter and then an order and limit, right? You take an animal quoter, you confess the picture, you use an norn end for the next state.

And these things were, you know, was a laugh at the bad simulator. But this is twenty fifteen error machine learning technology today. We have V Q V I E and transformers. We're building drive GPT basically .

have GPT O K A. So and is trained on what is IT trained in a supervise way.

It's trained on all the driving data predict next frame.

So really trying to what learn a human policy.

what would the human do well actually are similar condition on the post. So it's it's actually simulator. You can put in like a state action pair and get up the next state OK. Um and then once you have a simulator, you can do R L in the simulator. And R L will get .

us that .

human policy. R S with a reward function. Not asking is the closed to human policy, but asking what a human disengage if you did this behavior.

Okay, let me think about the the distinction there. What a human disengage, what a human disengage. That correlates, I guess, with the human policy. But IT could be so IT. IT doesn't just say what would a human do is is what would a good human driver do and such that experience is comfortable but also not annoying and net like the thing is very cautious finding a nice baLances that's interesting in nice.

It's asking exactly the right question, what will make our customers happy, right? A system that you never wanted .

disengage because usually this engagement is is almost always sign of. I'm not happy with all the system is doing usually .

are there some that just I felt like driving and those are always fine too, but they're just going to look like noise in the data.

but even they felt like driving map. Yeah, that's even that's a signal. Like why do you feel like driving here? You need to, uh, calibrate your relationship of the car. okay. So that's really interesting how closer we just solving self driving.

It's hard to say we haven't completely closed the loop yet. So we don't have anything built that truly looks like that. Architects, we have prototypes and the bugs. Um so we are a couple bug fixes away. Might take a year, my take .

time ah with the nature of the bugs are these are .

these major philopater gs. Um we just massively expanded our computer cluster. I come a, uh, we now have about two people worth compute. Forty.

bad of what? people? People are different.

Twenty, a person. I just the unit, right? Horses are different, but we still call on a horse power.

yeah. But there's something different about mobility and nurse about perception, action in a very complicated world. But yes, course.

not all flops are created equal if you have renal initial weight.

Not now, not all flaps are .

created equal. Possible things? OS.

no. Yeah, tell me about IT. okay. So more data scale means more scale computer, or scale and scale of data.

both.

Diversity of data.

Diverse is very important data. Uh the I mean we have so we have about think we have like five thousand daily activities.

How do you valuate? How F S D. Don't driving? How's that race going between test wanted two years ahead us?

They're always been one to two years ahead us, and they probably always will be because they not do anything wrong.

What have you seen that since the last time we talked? They are interesting architectural decisions, training decisions like the way the way they deployed off the architectures they're using in terms of the software, how the teams are around and of data collection.

I mean, I know they are moving toward more than approach.

so creeping towards end to end as much as possible across the whole thing, the training, the data collection, everything.

They also have a very fancy simulator. They're probably saying all the same things we are. They're probably saying we just need to optimize you know, what is the reward but your negative reward for the engagement? I like everyone kind of knows this is just question who can actually build deploy the system?

Yeah I mean quires sort engineering I think and in the right hand hardware .

yeah hard work.

You still don't believe in cloud in that regard.

Have a computer cluster in lia bus eight with body killed at idle or data center does in crazy before he killed the, I just burning just when the computer is, are I sorry, sorry. Computer cluster .

compulsory? I IT not .

date center. yeah. No data center are clouds. You don't have clouds did is always have air condition as we have fans that makes IT a computer cluster.

I'm guessing this is a kind of a illegal distinction.

sure. Yeah, we have a computer cluster.

You said that you don't think lumps of consciousness, or least not more than chicken, do you think they can? Reason is there is something interesting to about the word reason about some of the capital lives that we think is kind of human to be able to um integrate complicated information and through a chain of thought, arrive at a conclusion that feels novel a novel integration of the of desperate facts yeah I don't .

think that there's I think that can reason Better than a lot of people.

IT is not amazing to you though, is not like an incredible thing that a transform can achieve.

I mean, I think the calculators can add Better than a lot of people.

But language feels like reasoning through the process of language, which looks a lot like .

thought making brilliant season chess, which feels a lot like that whatever new thing that A I can do, everybody thinks is brilliant. And then like twenty years go by and we have a chest this like mechanical, like .

change language.

And because .

it's very human, we take IT, uh, we listen. There is something different between Jason and language. Chess is a game that a subset of population plays.

Language is something we use nonstop for all of our human interaction and human interactions fundamental to society. So how you should this this language thing is not so difficult to, like, create by the machine. The problem is.

if you go back to one thousand nine hundred and sixty and you tell them that you have a machine that can play amazing chess. Of course, someone in one thousand nine hundred and sixty, we will tell you that machine is intelligent. Someone in twenty ten world what's changed right today? We think that these machines that have language are intelligent. But I think in twenty years we're going to peak like, yeah but can IT .

reproduce so reproduction yeah we might redefine what he means to to be um what is that a high performance living organism on earth.

Humans are always going to define and dish for themselves like, well, you know we're Better machines because we can you know like they try creative for a bit but no one believes that on anymore .

what nation is is that is that delusional? Or is that some accuracy to that? Because maybe like with chest you start to realize that um we have ill conceived notions, what what makes you a special like the apex organism earth?

Yeah I think maybe we're gonna goes with that same thing with language, have the same thing with creativity.

The language Carries these notions of truth and saw on so we might be like weight. Maybe truth is not Carried by language. Maybe there's a deeper thing.

The dish is getting smaller.

a boy.

But no, no, no, you don't understand. Humans are created by god, and machines are created by humans. Therefore, right? Like that will be the last year here.

So what do you think about this? The rapid development, development, if we could just like, stick on that, is still incredibly impressive. Like a tragic ity, just even tragic ity. What are you thoughts about enforcement learning with human feedback on these large language models?

I'd like to go back to when calculators first came out and or computers and like, I wasn't around, look on on thirty three years old and to like, see how that affected. Like society.

maybe you right though I wanna put on the the the big picture had here. I got the .

figurated well, they're figurated .

electricity, all that kind of. But with the internet large language models seeing human like basic capacity touring test IT seems you might have really, at scale, rapid, transformative effects on society. We are saying, like other technologies have as well. So maybe calculators, not the best example that cause that just seems like, well, no, maybe calculate the milk.

Man, the day he learned about refrigerator is, have done. You tell me you can just keep the milk in your house. You don't need to deliver IT every day i'm done.

We that you have to actually look at the practical impact of certain technologies that they've had. You have probably electricity is a big one and also have rapidly spread. The thing is a big one.

I do think it's different.

Yeah feels like is .

getting smaller.

They initiated humans that makes humans special IT. Feels like it's getting smaller rapidly though, doesn't IT or is just a feeling. We dramatize everything.

I think we go out as everything. I think that that he asked the milk man when he saw vigiling ATS and they're going to have one of these in every home.

Yeah, yeah, yeah, yeah. make. But boys are impressive. So much more impressive than seeing A A chess world champion AI system.

I disagree. Actually, I disagree. I think things like new zero and alphabet are so much more impressive because these things are playing beyond and the highest human level. The language models are writing middle school level essay in people like Opera s uses .

the war just generating code cotex. So you're think it's media code ride, but I don't think it's there. I think it's mediocre code often close the correct like for media re just .

scare is kinda de. I spent five percent time typing and ninety five percent of time debugging. The last thing I want is close to correct code. I want a machine that .

can help me with the debugging ping like a level to a driving similar kind of thing. You still so should be a good programmer in order to modify. I wouldn't see the bargain. She's modifying the code reading IT don't .

think it's like level to driving. I think drivers is not tall complete in program this meaning you don't use like the best possible tools to drive. But you not you're not like, like, like cars have basically the same interface for the last fifty years. Yeah, computers have erratically different.

What can you describe the concept of tool complete? So think about .

the difference during a car from one thousand nine hundred and eighty and a car from today. Ight, no difference. Really got a bunch peddles guys during, well, great. Maybe now IT has a few ads features, but it's pretty the same car. You have no problem getting into a one thousand nine hundred eighty car and driving.

You take a programmer today who spent their whole life doing java script T, T, and you put him in an apple two e prompt, and you tell him about the line numbers in basic. But how do I insert something between nine, seventeen and eighteen? wow.

Uh, but so you in tour you're putting in the programing languages, so just the entirety stack of the tooling. So it's not just like the like ideas.

something like that, everything, the language, the one time program is is tool complete. So like almost if if if if if codex or copilot helping you, that actually probably means that your ramework library is bad and there's too much.

Boy, I played in IT.

Yeah.

but don't you think so much programme has always played?

Tiny grad is now twenty seven hundred lines, and IT can run lama and stable. And all of this stuff is in two thousand thousand seven hundred lines, boiler played and abstraction directions. And all these things are just bad code.

Ww, let's talk about good code and bad code, as I would say. I don't know for genre scripts that right just off hand like I like eighty percent of IT is written by GPT, just a quick, quick like off hand stuff. So not like libraries is not like performer code, not stuff for robotics and so on, just quick stuff because your basic so much of programming is doing some some yeah boller play. But to do so efficiently and quickly, because you can't really automated fully with like generic method, like a generic kind of um I D type of recommendations, something that you do need to have some of the complexity of language models.

Yeah, I guess if I was really writing like maybe today, if I wrote like a lot of like data passing stuff, yeah, I may. I don't play C, T, F anymore, but if I still play C, T, F, A lot of life is just like you have to write like a part data, like I wonder, or like add of code. I wonder when the models are going to start to help with that kind of code.

And they may, they may and the models also may help you with speed yeah and the model very fast. But where the models won't, I my programing speed is not at all limited by my typing speed. And in very few cases, IT is yes, if i'm writing some script to just like part some wear data format, sure, my programme is limited by my typing speed.

what looking stuff up because that's essentially a more efficient look up.

right? You know who knows that knows that twitter I try to use uh chat P T to uh to like asking questions like was the A P I for this and IT would just loosen IT would just give me completely made up A P I functions that sounded real.

What do you think that just a temporary kind of stage, you don't think you will get Better, Better, Better this kind of stuff because that only allusive a stuff in the h cases? yes. If if you're code is actually pretty good.

yes, if you are writing an absolute basic like react up with a button is a little nature. No, there there's kind of ways to fix a lucena problem I think faced because interesting paper has called at less. And it's actually weird the way that we do uh, language models right now where all of the uh, information is in the weights and human embrace.

Not really like this. We've like a hypoxia s in a memory system. So why don't l ms.

Have a memory system? And those people working on them, I think future allies are gna be like smaller, but are going to run looping on themselves and are going to have reproval systems. And the thing about using a retrieval system is you can find .

sources explicit, which is a really helpful to integrate the humanity to the loop of the of the thing, because you can go check the sources and you can investigate. So whenever the things is pollution inside, you can, like, have the human supervision that's pushing .

at our level to go .

go we which part.

when someone makes an ala lam that's capable of citing its sources, IT will kill google.

I am exciting the sources because it's basically.

but also google .

might be the people that build that. Maybe I put that on that.

I count them out.

Why is that? What do you think? Who who wins the race? We got who?

Who are the competitors? We got tiny corp. I don't know that mean your general computer in that i'm .

not trying to compete on that. You not no.

not just stumbled into that competition. You think you might build the search engine, replace google search.

When I start to come A, I said over and over again, i'm going to win self driving cars. I still believe that I have never said i'm going to win search with the tiny group, and i'm never going to say that because .

I want then I is still Young. Don't you don't know how hard is IT to win search in this new world like IT IT feels? I mean, one of things that ChatGPT kind of shows that there could be a few interesting tricks. They really have to create a really compelling product.

Some startups going to figure that out. I think. I think if you ask me, like google stole the number one, one page, I think by the end of the decay, google, will people want that anymore?

So you don't think google, because of how big the corporation is.

Look, I would put a lot more money on mark ebook. What is that? Because mark ara bera life like this is all program as I IT start up side there, alive or dead.

google's dead face versus facebook, alive.

alive. See what I mean. Like that's just like, like, like mark oker. This is mark occur. Reading that program missing and being like, i'm going to show everyone how alive i'm going to change me.

So you don't think there is a god pivoting engine that like google doesn't have that the kind of venture that the start up has like constantly you what being alive.

I guess when I lost your same altman podcast um he talked about the button everyone who talks about I talk button, the button to turn IT off. Friend, do you have a button to turn off google? Is anybody in the world capable of shutting google down?

What does that mean that .

either can you library on the value of that .

question does sound .

appreciate. Have the authority to turn off google outcome tomorrow.

Who has this special question?

Anyone.

anyone, I am sure.

Are you sure? No, they have the technical power. But do they have the authority?

Let's say, and appreciate made, this is so nation and came into google tomorrow. And I am gna shut google account down. Yeah I keep this position too long. And what is the mechanism .

by which he wouldn't keep this position? Well.

the boards and shares and corporate undermining and oh my god, our revenue is zero now.

okay. So what I mean, what's the case you're making here? So the the capitalist machine prevents you from having the button and IT will .

have mean the statue of the air too, right? There's not turning me a is off. There's no bottom. You can press IT. Now, does mark doka work have that button for facebook 点 com？

Yes, probable.

I think he does. I think he does. And this is exactly what I mean and why I bet on him so much more than I bet on google.

I guess you you .

stuff or you want has .

the button .

and does he can? He can fire the missiles, can he fire the misses?

I think some questions are Better. I asked.

right? I mean, you know a rocket and I C P, M, you a rocket that can land anywhere that an icp m ah you know, just to me .

questions, my god. Um but the positive side of the button is that you can innovate aggression what you say, which is what's required with the turning LLM into a search engine.

I bet on to start up.

I bet this is so easy.

I on something that looks like my journey, but for research .

just is able to set source of loop on. So that means just feels like one model can take off, right? And nice wrapper in some of IT is is hard to like create a product that just works really nicely stable.

The other thing that's gona be cool is there is some aspect of winner take all effect, right? Like once um someone starts to point a product that gets a lot of usage and you see this with opening eye, uh they're going to get the data set to train future versions of the model. Yeah and they are going to be able to you know, I was actually a google lime search, and I work there was fifteen years ago.

Now how does google know which image is an apple? And I said the meddle a and then like, yeah, that works about at the time. How does google know you will see the role applies on the front page when you search apple, and I don't come up with the answer human.

What do you think are the chances? What do you think in general that lama was open source? I just the the conversation with a with mark zao burgan and he's all in an open .

source who would have thought that mark zuker burg would be the good guy mean.

could i've thought anything in this world it's hard to know but open source to you util. Is a good thing here.

You know, what ironic about all these A I safety people is they are going to build the exact thing. They fear these. We need to have one model that we control in a line.

This is the only way you end up paper clipped. There's no way you end up paper collective. Everybody has an A I.

So open sourcing is the way to fight the paper clip max solution.

The only way you think you're gonna control IT, you're not gonna control IT.

So the criticism you have for the A I A T folks is that there is a belief in a desire for control there, and that belief and desire for centralized control of dangerous A I systems is not good.

Sam won't tell you that G P T four has two hundred twenty billion parameters and is a sixteen way mixture model with eight sets of weights.

Who who did you have to murder to get that information? I mean.

everyone to opening, I knows what I just always true right now. Uh, ask the question really. You know, IT upsets me when I like GPT two. When opening, I came out with GPT two and raised a whole fake A I safety thing about that. I mean, now the models level like they they used A I safety to hide up their company and it's disgusting.

Or the fool s side of that is they used a relatively weak model in retrospect to to explore how do we do A S, C, T correctly, how to release things, how do we go through the process. I don't, I don't know.

I don't, I don't know how much hy, I don't .

know how much .

hype is in a come .

on in terms of type. I mean, I don't I I think company, I have been finding interesting baLance between transparency and putting value on a you don't think you think just go all out open source to do a lama to do like open source this this is a tough question, which is open source both the the base, the foundation model and the fine tune one. So like that, the model that can be ulto racist and dangerous, like, tell you how to build a nuclear weapon.

God, have you met humans? right? Like, have that is A I I haven't .

met most humans. I this makes, this allows you to meet every human. Yeah.

I know. But how of these? A I alignment problems are just human alignment problems. And that was also so scary about the language they use IT. Like it's not the machines you wanna .

line to me but here's the thing IT makes IT very accessible .

to ask .

very questions where the answers have dangerous consequences if you were to act on them I mean.

yeah welcome to the world but .

not for me. There's a lot of friction if I want to find out how to um I don't know, blow up .

something not .

like what you search these .

or just no.

to keep anyone who stupid enough .

to search for how to blow up a building in my nights or hood is not long enough to build bob, right?

I sure.

Yes.

I I feel like, I feel like a language model makes IT more accessible for that person who is not smart .

enough to do a bomb. Trust me, the people, the people who are incapable of figuring out how to, like, ask that question a bit more academically and get a real answer from IT, are not capable of procuring the materials, which are somewhat .

control to no, I think, makes you more accessible to people with money, without a technical know how right to build. Like you do really need to know how to build a bomb. To build a bomb, you can hire people .

you can find hire people to build. You know what? I was asking this question on my dream. Like, can jeff BIOS higher hit man.

probably, but a language model can probably help you out.

Yeah, you'll still go to jail, right? Like, it's not like the language model is got like the language model. It's like it's literally just hire someone on K.

G, P, T four. In terms of finding hit man is like asking five or how to find A, I understand.

But don't you think .

don't you think GPT five will be Better? please? Do you think that information is out .

on the interview? I mean, yeah, and I think that if someone is actually serious enough to hire a hit man or build a bomb, theyd also be seriously ough to find the information.

I don't think so. I think that makes them more accessible if you have if you have enough money to buy him, man, I think the decrease is the friction of how hard is IT to find the kind of him. I honestly, you think this there's a jump in ease and scale of how much harm you can do. And I don't mean harm with language. I mean harm with actual violence.

What you're basically saying is like, okay, what's happiness? These people who are not intelligent are going to use machines to augment their intelligence. And now intelligent people and machines intelligence scary.

Intelligent agents are scary. When I of the woods, the scary is the animal to me meet is human, right? No, no. This look is like nice california humans. Like I see you're wearing like, you know, street clothes and nike are are fine, but you look like you're been a human being in the woods for a while. Yeah, i'm more scared to you than a bear.

And amazon mazo human tries.

So intelligence is scary, right? So lect ask a question generic, where you're like, what if we took everybody who, you know maybe has a ill intention, but is not so intelligent and gave them intelligence? right? So we should have intelligence control. Of course, we should only give intelligence to good people. And that is the absolutely half an idea.

Should you best defenses, you best fence to give more intelligence? Good, good. And give intelligence.

Intelligent body know, even like gun, like people say about, you know what what's the best defence against bag of A N I subscribed to that, but I really subscribe that with .

intelligence yeah, in a fundament way I I agree with you, but there's just feels like so much uncertainty and songs can happen rapidly that you can lose a lot of control. You can do a lot of damage.

Oh, no, we can lose control. Yes, thank god. yeah. I hope, I hope they lose control. I want them to lose control more than anything else.

I think when you lose control, you can do a lot of damage. We can do more damage when you centralized and hold on to control is the point .

of centralized and held control is tyranni will always unlike anarch either. But I always take an active tony. And can I have a chance .

this human societies will get going on? It's quite interesting. I mean, I agree with you. So do you open source is the way forward here so you admire what facebooks is doing here and what matter is doing with at least them?

Yes, I was. I was eighty thousand dollars last year investing in meta. And when they released lama, i'm like, yeah, whatever meant was worth.

worth IT. Do you think google and the OpenAI old microsoft match what what what matters doing? You know?

So if I were a researcher, why would you want to work at all my eye? Like you know you're just you're on the bad team like I mean, I like you're on the bad team who can even say the GPT four has two hundred twenty billion .

private so close source use the bad team.

I'm only close source. I'm not saying you need to make your model weight open. I'm not saying that I totally understand we're keeping our model way to close because that's our product and not fine. I'm saying like because of A I safety reasons, we can't tell you the number of billions of parameters in the model. That's just the bad guys.

Just because you're mocking eyes. Safety doesn't mean it's not real. Cause is impossible. These things can mage.

oh god, intelligences dangerous. Human intelligence or machine intelligence. Intelligence is dangerous.

Machine intelligence is so much easier to deploy the scale, like rapidly. Like what okay, if you have human like .

boss on twitter and .

you have like a thousand of them, create a whole narrative like you can manipulate millions of people.

But you mean like the intelligence amErica are doing right now?

Yeah but you're not doing IT that that well, IT feels .

like you can do a lot. I think everyone a prety good job.

I I suspect in I nearly as good as a bunch of GPT field boss could be. I mean, of course.

they are looking into the latest technologies for control of people, of course.

But I think that is George house type character that can do a Better job that need tirely of them. You don't seeks why?

no. And i'll tell you why the George hot character can. And I thought about this a lot with hacking, like I can find x in my problem.

I probably still am I Better twenty? But if you think that I lack is the ability to slowly and steadily deploy them over five years. And this is what intelligence agencies are very good at, right? Intel's ence agrees. Don't have the most sophisticated technology. They just have injuries, injuries.

And yeah the financial backing and the infrastructure for the experience.

So the more we can decentralize power, you like you can make an argument, by the way, that nobody should have these things. And I would defend that argument. I would I would like you saying that, look, elms and A I and machine intelligence can cause a lot of harm, so nobody should have IT.

And I will respect someone philosophically with that position, just like I will respect someone philosophically with the position that nobody should have guns, right? But I will not respect philosophically, which is with with only the trusted authorities should have access to this. Who is the trusted authorities? You know what? I'm not worried about alignment between A I company and their machines. I'm worried about alignment between me .

and A I company. What do you think the either the cost would say to this is really against open source?

I know. and. I thought about this, I thought about this, and I think this comes down to a repeated misunderstanding of political power by the rationalists.

Interesting.

I think that alli's uci is scared of these things. And I am scared of these things too. Everyone should be scared of these things.

These things are scary. But now you ask about the two possible futures. One where a small, trusted, centralized group of people has them, and the other, everyone has them. And I am much less scared of the second future. In the first.

well, there's a small trusted group of people that have control of our nuclear weapons.

There's a difference. Again, a nuclear weapon cannot be deployed technically, and a nuclear weapon is not a defense against nuclear weapon, except maybe in some philosophical mind game kind of way.

But A I is different. How exactly?

But let's say the intelligence agency deploys a million bots on twitter, or thousand bottles on twitter to try to convince me of a point. Imagine I had a powerful eye running on my computer saying, okay, and nice sign up. Now sign up.

Now sign up. Okay, just sign up. I felt to do that for you.

yeah. I mean, do you have fundamental hope for that for the for the defensive side up?

Even like I don't even mean these things are like truly horrible ways. I mean these things and straight up like blocker, right? Yes, sure, bad locker I don't want, but they're always finding, you know imagine I had an A I that could just block all the ads for me.

He believe in the the power of the people. Those create a now blocker ah I mean I I can't share that belief. I have that one of the deepest optimisms I have is just like there's a lot of good guys so to give you so you should and hand pick them, just throw out powerful technology out there and the good guys will out number and out power bad guys yeah not going to say .

there's a lot of good guys. I'm saying that good out numbers. Bad right? Good out numbers.

Bad in skill and performs.

Yeah definitely in skills and performance. Probably just a number two, probably just in generally me if you know you believe philosopher and democracy, you obviously ve that um that good and bad and like the only if you give in to a small number of people, there's a chance you gave IT to good people, but there's also a chance you give IT to bad people. If you give IT to everybody with good out numbers, bad, then you definitely gave IT to more good people than bad.

That's really interesting. So that sound the safety grounds. But then also, of course, there's other motivations like you don't want to give away your secret sauce.

That's I mean, I am. I I look, I respect italy. M, I don't think that. I think that I would be polite for you to make model architectures open source and fundamental breakthrough open source. I don't think you would .

have to make weight open course. You know, this interesting is that, like there's so many possible trajectories in human history, where are you could have the next google be open source? So for example, I don't know that connection is accurate, but no, kip dia made a lot interesting decisions not to put ads like wikipedia basically open source.

You could think of IT that way. yes. And like that's one of the main websites on the and I didn't have to be that way.

You could have been like google cut of credit, wikipedia put ads on IT. You could probably run amazing as now kip dia, you wouldn't have to keep asking for money, but it's interesting, right? So lama, open source, lama, the rivers open source law might win the internet.

I sure hope so. I hope to see another era. You know, the kids today don't know how good the internet used to be, and I don't think this is just come on like everyone's nosti c the past, but I actually think the internet, before small groups of weapons, zed corporate government interests took IT over, was a beautiful place.

You know those small number of companies have created some sexy products, but you're saying overall, in the long art of history, the centralization of power they have like suffocated the human spirit scale. Here's a question .

to ask about those beautiful sexy products. Imagine two thousand google to twenty ten google, right? A lot changed. We ve got maps, we got gmail.

We lost a lot of products. So I think from here.

I mean, somewhere problem, you get chrome, right? And now let's go from twenty ten. We're got in dropped.

Now let's go from twenty ten to twenty twenty. What does google have to change? Maps, ml, android and chrome? 啊。 I see the internet. Was this, you know, I was time first one of the year two thousand six. Yeah.

I love this .

two thousand years like that. H, you know, so quickly did people forget? And I think some of its social media, I think some of that I I hope, look, I hope that I I know, if possible, that some very senior things happened. I I don't know. I think I might just be like the effects of social dia, but something happened in the last twenty years.

No, okay. Is just being an old man is worry about. I think there is always IT goes. It's the cycle thing is also downs. And I think people rediscovered the power of distributed, of decentralized. Yeah I mean, that's kind of like what the whole critter mercy trying to exist that I think crypto is just Carrying the flame with that spirit of likes, soft should be essential.

Just just such a shame that they all got rich, you know? Yeah, if you took all the money at a script out, or would have been a beautiful place, yes, but now, I mean, these people, you know, they sucked all the value out of IT and took IT. Yeah.

money kind of corrupt the mind. somehow. IT becomes the drug.

corrupted all of crypto. O, you had coins worth billions of dollars that had zero use.

You still have hope for capital?

sure. I have hope the ideas I really do. Um yeah I mean, you know. I want your dollar to collapse. I do George's well.

let me sort of on on the A S city. Do you think there's some interesting questions there though to solve the open source community in this case? So like alignment, for example um or the control problem, if you really have super powerful, he said that scary ah what do we do with that?

So not not control, not in face control. But like if you were, then you're gonna see some guy or gal release a super powerful language model. Open source.

Here you are, George hosking, holy shit. Okay, what ideas do I have to combat this thing? So what ideas would you have?

I am so much not worried about the machine independently doing hard. That's what some of these A I safety people seem to think. They somehow seem to think that the machine, like independently, is gonna rebel against its creator.

So you don't think you'll find a tony.

No, this is sif I B movie garbage.

okay? What if the thing writes code? This right viruses.

If the thing that right viruses, it's because the human told IT the right viruses, but there's .

some things you can like put back in the box is so kind of whole point is a kind of spreads give existing international spreads, installs .

itself.

modifies not real. But i'm trying to work. I'm trying to get Better in .

my plot writing the the thing that worries me. I mean, we have a real danger to discuss and is bad humans using the thing to do whatever bad on the line I think you want.

But this goes to the, uh your previous concern that who gets define who is a good human, who is a bad human.

Nobody does. We give IT everybody. And if you do anything besides give IT to everybody, trust me, the bad humans will get IT. Is that to get power? Always the bad humans to get power.

okay, power and power turns even slightly good humans to bad. That's the intuition you have. I don't know. I don't .

think everyone. I don't think everyone. I just think that like here, here's the saying that I put in one of my blog posts.

When I was in the hacking world, I found ninety five percent of people to be good, in five percent of people to be bad. Just why personally, judges is good people and bad people, like, they believed about, like, you know, good things for the world. They wanted, like flashing, and they wanted, you know, growth, and they wanted, think, I can say good, right?

I came into the business world with comma, and I found the exact opposite. I found five percent of people good and ninety five percent of people bad. I found a world that promotes psychopathy.

I wonder what that means. I wonder if I can like, I wonder that dollar, if IT, if that's true, that there's something about capitalism at the core that promotes the people that run capitalism, that promotes the company.

That saying may, of course, be my own biases, right? That maybe my own biases is that these people are a lot more aligned with me than .

these other people .

yeah so you know, I can certainly recognize that. But you know, in general, I mean, this is like common sense maxim, which is the people who end up getting power or never the ones you want with that.

But do you have a concern of super intelligent A G I open source? And then what do you do with that? I'm not saying control IT. It's open source. What do we do with these human species that's not to meet?

I mean, you know like i'm not a central planner.

not central planner, but you probably tweet as a few days left to live for the human space.

I have my ideas of what to do with them, and everyone else is their ideas of what to do and the best ideas when.

But at this point, do you brain like because it's not regulation, IT could be decentralized regulation where people agree that this is just like we create tools and make IT more difficult for you to maybe make IT more difficult for code to spread, you know, anti virus software, this kind of thing. But saying that you should .

build a ee firewall, that sounds good, you should definitely be running an A I firewall. You be run an A I firewall to your mind, right? You're constantly .

under the .

interesting IT .

is a info war is man like, I don't know, I C I power. How do I protect my mind from influence of human leg or superhuman intellect?

Bots, I not being. I would pay so much money for that product. I would pay so much money for that product. I would get so much money. I paid just for a spam filter that works well.

And twitter, sometimes I would like to have A A protection mechanism for my mind from the outrage mobs because I feel like bot like behavior is a is a large number of people that will just grab a vial narrative and attacked anyone else to believe othe wise. And it's like whenever .

someone's telling me some story from the news, I always like, I don't want to hear A C I I. Opera is a cii Opera like doesn't matter that's true or not. It's just trying to influence your mind. You're repeating an add to me the viral modes of this is is like .

they are there there to me IT defense against those those mobs is just getting a multiple prospect is always from from sources that make you feel kind like you getting smarter and just actually just basically feels good, like a good documentary just feels something feels good about IT is well done like, okay, I never thought of IT this way. Feels good sometimes the outrage mobs, even if they ever could point behind IT when they're like marking the receive and just aggressive, you're with us, against us this and .

this is why I delete my tweet.

Yeah, what you do that, you know, as I miss your tweet, know what IT is?

The algorithm promotes toxic?

no. And like, you know.

I think elon has a much Better chance of fixing IT than the previous regime.

yeah.

But to solve this problem, to solve ability, social network that is actually not toxic without moderation.

like not to think book care, so like what people, uh, look for goodness. So that would make IT catalyze the process of connecting cool people being called to each other.

Yeah.

without ever sensory, without ever sensory.

And and and like Scott sander has a blog poster like we talking about, like moderation is not censorship, right? Like all moderation you wanted put on twitter, right? Like you could totally make this moderation.

Like just a you don't have to block IT for everybody. You can just have like a filter of button might that people can turn off if they would like, say, search to her, right? Like some good to turn that off, right? So like, but then you'd like, take this idea to an extreme, right.

Well, the network, you just show you, this is a couch surfing C E O thing, right? If IT shows you right now, these algorithms are designed to maxim engagement turns out out range maxes engagement quark of human, quark of the human mind, right? Just I falls for IT have one false for IT um so yeah, you got ta figure out the max with something of the engagement .

and I actually believe you can make money with that too. So it's I engaged is the only way to make money.

I actually think it's incredible that we're starting to see I think again, you want to doing so much stuff right with twitter like charging people money as soon as you charge people money than no longer the product, the customer and then they can start building something that's good for the customer and not good for the other customer, which is the .

other as isn't picked up steam.

I pay for twitter doesn't get me anything. It's my donation to this new business model. Hofus ly working out.

sure. But you know you for this business model to work is like most people should be signed up to twitter. And so the way was there was something I perhaps not compelling or something this stood to.

People think you need most people at all. I think that why do I need most people, right? I don't make an eight thousand person company, make a person company.

Well so speaking of which ah you worked at twitter for a bit as an intern, the worlds greatest turn. That's been Better. Tell me about your time at twitter. How did you come about and what did you learn from the experience?

So I deleted my first twitter in twenty ten, had one hundred thousand followers back when that actually meant something. And I just saw up, you know, my coker summarized IT well, he's like, whenever I see someone twitter page, I either think the same of them or less of them. I never think more of them. I I to mention any names, but like some people who like, you know, maybe you would like read their books and you would respect them. You see them on twitter and you're like, okay, dude.

yeah but there are some people would say, know who I respect a lot are people that just post really .

good technical stuff .

and I guess I don't know. I think I respect more for IT cause you you realized all this wasn't the there's like so much death to this person, to their techno understanding so many different topics. okay. So I try to follow people. I tried to consume stuff, that technical machine learning content.

There's probably a few of those people. And the problem is inherently what the algorithm rewards, right? And people think about these algorithms.

People think that they are terrible of of things. And I ve sce um because I mean, what he does is actually pretty obvious. IT just predicts what you are likely to retreat and like and linger on.

So all these adventures, but tiktok is. So all these recommendation and instill and IT turns out that the thing that you are most likely interact with his, our age, and that's a quick at the human condition. I mean, any different .

laws about age IT doesn't have to be IT could be mockery. You be in the topic of our age could be different that could be an idea that could be a person that could be maybe that's about a word in our age could be drama .

sure stuff no but doesn't feel .

like when you consume IT is a constructive thing for the individuals that consumer in a long term.

Yeah so my time there I absolutely couldn't believe, you know I got crazy amount hey uh you know, on twitter for working a twitter IT seems like people associated with this, I think maybe .

were posed to the twitter .

and d laon like the whole .

elan got ten a bit spicy during that time, bit political a bit. Yeah yeah.

You know, I remember one of my tweet was never go full republican and he long like IT.

Boy, yes. I mean, there's a roll cuter that was being political on twitter.

Yeah boy, yeah. You're also .

been just attacking anybody on twitter. IT comes back at you harder. And if his political .

and attacks are absolutely .

and then letting ah sort of the platform people back on even adds more fun to the to the to the beautiful chaos I .

was hoping. And like I remember when you want talk about buying twitter, like six months earlier, he was talking about like a principled uh, commitment for speech and i'm a big believer in fan of that I would love to see an actual principal commitment to free speech um of course this isn't quite what happened um instead of the oligarchy deciding what to ban, you had a monarch deciding what to ban, right? Instead of all the twitter file shadow, really the work, you just decide what cloth masks are productive against cov IT. That's a true statement every doctor and two and nineteen new year and now I ban on twitter for saying IT interesting all guy, i'm so now you have a monarchy and you know Evans and things he doesn't like so you know, it's just just a different it's different power and like, you know, maybe I, maybe I alive more with .

him than with the collegiate but is not but I I feel like being a face peach. Absolutes on the social network requires you to also have tools for the individuals to control what they consume easier like, not sensor, yes, but just like a control like i'd like to see more cats and less politics.

And this isn't even this isn't even remotely controversial.

This is just saying you want .

to give paying customers for a product what they want, not through the lock. Individually .

transparent. Sorry, right?

yeah. But censorship is a strong word, and people are very sensitive to fan. When I look at right now.

i'm looking at you, i'm sensors, everything else out when i'm when my mind is folks at you, you can use the word censorship that way. But usually when people get a very sensitive about the censorship thing, I I think when you have when anyone is allowed to say anything, you should probably have tools that maximize the quality of the experience for individuals.

So, you know, for me, like what I really value, boy would be amazing to somehow figure how to do that. I love disagreement and debate, and people who disagree with each other disagree with me, especially the spaces of ideas, but the high quality ones. So not the reason, right?

Probably there's just .

the way of talking the narky and so on that somehow is gets people on twitter and they get excited and so on.

We have like at home and refusing the central point.

I like think this is an central pair. yes. And I like all of IT, all the wrong stuff is attractive to people.

I mean, we can just train to classify to absolutely say, what level of mathild hierarch of argument are you at? yes. Is ham like? Okay, cool. I turned on the no ad homine filter.

I wonder if there's A A social network that will allow you to .

have that kind of filter. Yeah so here's a problem with that. Um it's not going to win in a free market. What wins in a free market is all television today is a reality television because is engaging, right? If if if engaging is what wins in a free market, right? So IT becomes hard to keep these other more no wth values.

Well, okay. So that's the experience of being on twitter. But then you ve got a chance also together with the other engineers and with the launch of look, brain storm, when you step into a code base has been around for a long time.

You know, there's other social networks, facebook, this is old code basis instead. And see, okay, how do we make with fresh mind progress on this cold place? Like what you learn about soft engineering, about programing? Just experience .

my technical recommendation, and I said this on the twitter spaces afterward. I said this many times during my brief international p um was that you need to factors before features. Um this code base was I look, i've a goal. I watch a facebook um facebook has the best code ah, then google, then twitter um and you know what, you can know this because look at the machine learning frameworks like facebook release py orc h, google release TensorFlow and twitter released. Okay, so how you know what .

is is a proxy. But yeah that the google call is quite interesting. There's a lot of but the costs very large.

The code base was good in twenty in two thousand and five IT looks .

like two thousand five hours teams very difficult. Um I feel like twitter does less I obviously much less than google in terms of like the set of features, right? So like I can imagine as the number south engineers that could recreate twitters much smaller than to recreate google.

yeah, I still believe in the amount of hate I got for saying this, that fifty people could build a maintain .

water pretty .

nature comfortably.

But you don't know what you're talking.

You know what that is and it's the same. This is my summary of like the hate I get on hacker news. It's like when I say i'm going to do something, they have to believe that it's impossible because if doing things was possible, theyd have to do some soul searching and ask the question, why didn't .

they do anything? I people go like, what are your credentials? What to help you talking about? What is this is extremely difficult. Of course, you're new that to understand the problem deeply. I mean that those the same nature of hate, they're probably you on go when you first talked about the time was driving but you know, there's person concert is like, you know, there is experts in this .

world personed experts, the market, the people who are mocking are not experts with carefully reasoned arguments about why you need eight thousand people to run a bird APP there. But the people are gonna SE their jobs. Well.

that but also the problem says, no, it's a lot more complicated than you realize. But maybe he doesn't need to be so complicated.

You know, some people in the world like to create complexity. Some people in the world thrive under complexity like layers, right? Lawyers want the world to be more complex because you need more lawyers need more legal um and that's another if there's .

two great evils in the world at centralization and complexity yeah know the one of the hidden side effects of software engineering is like finding pleasure complexity I I don't remember just taking all the service engineer courses just doing programme in this is coming up in this uh uh object in a program kind of idea you don't like not often do people tell you like do the simplest possible thing like like a professor, a teacher is not going to get in front.

Like this is the simple way to do IT. They'll say like this is that there's the right way and the right way, at least for a long time. You know first I came with like java, right? Like is so much boiler plate, so much like so many classes, so many like designs in architecture and so on, like planning for features far into the future and planning poorly and all this kind of stuff.

And then there's this like code base that falls you along and put pressure. I knew nobody knows what like parts, different parts do, which slows everything down, is a kind of bureaucracy that instill in the code as a result. But then you feel like, oh, well, I follow goods often of engineering practices is initiating trade c because then you look at like the get owner of like perl and the old, like how quickly you can just write a couple ized you get i've done that trade off is interesting, or bash or whatever these kind of get all things you can do.

Linux, one of my favorite things to look at today is how much do you trust your test, right? We put a ton of effort in common, and I put a ton of effort, tiny grad, into making sure if you change the code and the tests pass, that you didn't break the code. Now, now, the obvious ly is not always true, but the closer that is to true, the more you trust your test, the more you're like, oh, I got a pull request and the test pass, I feel like the faster you can make .

people always programing to test in mind, developing tests that mind that if he passes IT should be good.

Twitter.

that not that so that was impossible .

to what other stuff .

can you say about the costs that made IT difficult? Um what is some interesting sort, of course, broad speaking from that compared to just your experience with 蛤蟆 and ever else？

The real thing that I I spoke to a bunch of you like individual contributor, a twitter, and I got to stand, stand like, okay, so like, what's wrong with this place? Why does this code look like this? And they explain to me what twitter's promotion system was, the way that you ve got promoted.

A twitter was you wrote a library that a lot of people used, right? So some guy wrote an engine replacement for twitter. Why does twitter need an engine replacement? What was wrong with engine x when you see you're not gonna get promoted if you use interacts, but if you write a replacement and lots of people start using IT as the twitter front end for their product, then you're going .

to a get promoted, right? So interesting is, like from the individual perspective, husson device, how do you create the kind of sentence that will lead to a lead to a great, great code base? Okay, what's the answer to that?

So what I do at coma and that, uh, you know, a tiny corp, is you have to explain IT to me after to explain me what's code does, right? And if I can sit there and come up with a symbolic way to do IT, you have to rewrite IT. You have to agree with me about the simple way, obvious conversation about this is a tory if you're like, ow, like that actually is what is simpler. Like, like this. Simplicity is important.

but that requires people that overlooked the code at the highest levels to be like.

okay, IT requires techno leadership.

be trust, yeah, technical leadership. So managers of whatever should have to have technical savi, deep technical vi manager .

should be Better programmer than the .

people who they manage yeah, and that's not always obvious. Al, to create a large companies, manager gets soft. And like.

you know, this is just I can still this culture at comma and coma has Better problem with to me who worked there but you know, again, i'm like to know the old guy in goodwill hunting. It's like luck, man. You know, I might not be as good as you, but I can see the difference.

Treat menu right? And yeah, this is what you need. This is what you need to the top. You don't necessarily need the manager to be the absolute best, say that. But like they need to be able to recognize scale.

They have good intuition. Intuition, that lady, with wisdom, for all the battles of trying to reduce complexity. Code basis.

You know, I took A I took a political approach coma to that. I think it's pretty interesting and thing you on the political approach. Google had no politics.

And what ended up happening is the absolute worst kind of politics to cover. Comma has an extreme amount of politics and they're all mine and no dissidents. S is tolerated.

So IT is a dictatorship.

an absolute dictatorship, right? You want does the same. Now, the thing about my dictatorship is here in my values.

Yeah, it's transparent. It's transparent.

It's a transparent dictor ship, right? And you can choose the option, you know, you get free access, right? That's a beauty of companies. If you don't like the data ship, you quit.

So you mentioned we right before or reactor before features. If you were to reference the twitter code base, what would that look like? And maybe also comment on how difficult to story factor.

The main thing I would do is first of all, identify the pieces and then put tests in between the pieces. right。 So there's all these different twitter ers of micros architecture. Um different. And the thing that I was working on that look like you, George, didn't know any java script. T he asked how to fix, search the man if the thing is like, I just, you know, i'm upset that the way that this whole thing was portrayed, because IT wasn't like, IT wasn't like taken by people like honestly, IT wasn't like by IT was taken by people who started out with a death assumption yeah and I mean, I look.

I can't like and use the program or just being transparent out there. Actually having like fun and like this is what I ve .

that me opportunity and know the day I quit came on my twitter spaces afterwards we had a conversation like.

I just I respect that so much yeah. And also inspiring engineers and programmer. It's cool if she'd fine the people they are hating on. IT is like, man, I was.

I was fun. IT was dressed but I felt you know was like a cool like point in history and like I hope I was useful and probably .

kind of wasn't one of the people that economy a strong case to the factor yeah and that's a really interesting thing to raise like maybe that is the right you know the timing of this was interesting. If you look at just the development of autopilot, you know, going from mobile I to just like more if you look at the history of simula was driving a tesla is is more and more like you can say we factory or or starting from scratch, redeveloping from scratch.

It's reactor.

although I down and like a and the question is that can you do that sooner? Can you maintain product profitability? And like what's what's the right time to do IT? How do you do IT you know, any one day? It's like, you don't want to pull off the bad days. I like everything works, is just like little fix here there, but maybe started from scratch.

This is the main philosopher, tiny grad. You have never reactor enough. Your code can get smaller. Your code can get simpler, your ideas can be more elegant.

But would you consider, you say you are like a running twitter development teams, engineering teams, would you go far like different programing language? Just go that far.

I mean, the first thing that I would do is bill tests. The first thing I would do is get A C I to where people can trust to make changes. So if you keep touched any code, I would actually say, no one touches any code. The first thing we do is we test this code in this classic. This is how you approach a legacy code, is like what how to approach a legacy code based book will tell you so.

And then you hope that there's modules that can live on for a while and then you add new ones may be a different language or before someone's .

we replace the old ones, yeah.

I mean, replace old ones with something simple.

We look at this, like this thing that one hundred thousand lines and we're like, well, okay, maybe this did even make sense ten, but now we can replace this with an open source thing, right? yeah. And you know, we look at this here, here's an other fifty thousand lines.

Well, actually, you know, we can replace this with three hundred lines ago. And you know what? I trust that the goal actually replaces this thing because all the test i'll pass.

So step one is testing now and then step two is like the programing languages and after thought, right, you know, let a whole lot of people compete. Be like, okay, who wants to rewrite a module? Whatever language want to write IT in just the test have to pass. And if you figure out how to make the test pass, but break the site that we got to go back to step one. Step one is get tests that you trust in order to make changes in the cops.

No harder is to because i'm i'm with you on on testing on everything have from tests to like a third to everything with codes just covering this because this should be very easy to make rapid changes. And no, that's not going to break everything and that's the way to do IT. But I I wonder how difficult is is to um integrate tests into a code basit doesn't have many of them.

Tell you what my plan was a twitter. It's actually similar to something music comments. So come on, we have the thing called process reply and we've a bunch of out that they will be run through.

So come as a micros architectural with microwaves in the driving. We have for the cameras, one for the sensor, one for the planner, uh, one for the model. And we have A A, P, I, which the microwave struck each other with.

We use this customer called syria, which uses M, Q. Twitter uses a thrift, and then IT uses this thing called the nail to the scala, uh, uh, R, B, C. back.

And but this isn't really matter. The thrift and finagle layer was a great place, I thought, to write tests. I just start building something that looks like process reply.

So twitter had some stuff that looks kind of like this. But IT wasn't offline, was only online. So you could ship like a modified version of IT and then you could redirect some of the traffic to your modified version and diff those too.

But IT was all online. IT was not like C. I. In the traditional sense. I mean, there was some.

but like IT was not full coverage. You can run all of twitter offline. something.

This was another problem. You can't run all of twitter right, period. Twitter, one person.

twitter runs in .

three data centers and that's IT. There's no other place you can run twitter, which is like, George, you don't understand this is modern software development. No, this is bullshit. Like I get to run on my laptop OK well, not saying you're going to down with the whole database. You up up i'm saying all the middle and the front and should run on my laptop, right?

That sounds really compelling. But can I be achieved by a code based a girls over the years? I mean that three data didn't have to be right because is there totally .

different like designs? The problem is more like like why did the code base have to grow? What new functionality has been added to compensate for the the lines of code that are there?

One of the west explain is that the incentive for softer developers to move up in the companies to add code to add, and especially .

in the for politicians to move up in the political structures to add laws, same problem.

Yeah yeah. If a the flip side is to simplify, simplify, simplify.

You know what? This is something that I do differently from from from elan with with comma about self driving cars. You know, I hear the new versions gonna come out and the new version is not going to be Better. But at first, and it's onna require tony y factors, I say, okay, take as long as you need like you convince me this architecture is Better, okay, we have to move to IT even if it's not going to make the product Better tomorrow. The top priority is making is getting the architecture right.

So what do you think about sort of a thing with the product is on mind. So how, I guess, would you do a reactor if you run engineer in twitter? Would you just do a reactor? How long would you take? What would that mean for the running of them of the actual service.

you know and. I'm not the right person to run twitter. I'm just not and that's the problem. Like like I don't really know.

I don't really know if that you know a common thing that I thought a lot while I was there, whenever I thought something that was different to what you want, thought i'd have to run something in the back of my head, reminding myself that iran is the richman in the world, and in general, his ideas are Better than mine. Now there's a few things I think I do understand and know more about. But like in general, i'm not call to around twitter, I say qualified.

But like I don't think i'd be that good at I don't think that be good at 了。 I don't think that really be good at running and engineering organization at scale. I think I could lead a very good reactor of twitter and IT would take like six months to a year.

And the results to share at the end of IT would be feature development in general. Takes ten x last time, ten last. That's what I think I could actually do. Um do I think that is the right decision for the business above my bag? D.

Yeah, but a lot of these counts of decisions are above everybody is pay grade.

I don't want to want to do that. I just like if you really forced ed me too, yeah, I would make me maybe. Make me sad if I do make those decisions, why do not want to?

Yeah, but a receptor is so compelling. If this is to become something much bigger than what twitter was is IT feels like a reactor has to becoming at some point George.

join yourselves for engineer. Every je yourselves for engineer wants to come in and reject all coding. Okay, like this, like your opinion. Man.

yeah. Doesn't you know sometimes they are right. Well, like whether they're right or not.

it's definitely after that reason, it's definitely not a question of engineering power. IT is a question of maybe the priorities the company and I did get more intelligent like feedback from people I think in good faith, like saying that um from actually million and like you know from from milan sorter like like people were like, well, you know, I stop the world ref actor might be great for engineering, but you know, business on and hey.

above my pay would you think about ilan is an engineering leader having to experience them in the most chaotic of spaces, I would say.

My respect for him is unchanged um and I did have to think a lot more deeply about some of the decisions he's forced .

to make about the attention that that knows the trade. Also those .

decisions about like a whole like like matrix coming at him and that undertakes word for IT sorry to borrow .

IT also bigger engineering just .

everything yeah like like the wall on .

the walk yeah .

like it's just it's just man and like he doesn't have to do this, you know he doesn't have to he could go like para and go chill the four seasons mani, you know, but see, one person I respect and one person I don't.

So his heart is in the right place fighting, in this case, for this idea of the freedom of expression.

But I wouldn't define the idea. So simply, I think you can define the idea no more, then just saying, elon's idea of a good world freedom of expression is but you.

So the downsides of that is the monarchy.

Yeah, I mean, monarchy has problems, right? But I mean, what I trade right now, the month of the current alga, which runs amErica for the monarchy, yeah, sure to the elon market. yeah.

You know why? Because power will cost once to get out out. Kind of sent a kilo hour.

What mean right now I pay about .

twenty cents a kilo, one hour for electricity. And 第二个， that's like the same Price you paid in one thousand nine hundred eighty.

So you would see a lot of innovation body on maybe .

maybe have some hypo loops.

Yeah right.

And i'm willing to make that trade off, right? I'm willing to make. And this is why you know people think that like dictators take power through some like through some untoward mechanism.

Sometimes they do, but usually they discuss the people want them and the downsides. Have a dictatorship. I feel like we've got to a point out the other garage where, yeah I would provide the dictator.

Ah what do you think about school is the programme language?

I like this more than I thought. I do the tutorials that I was very new to IT. Like I would take me six months to be able to write like google. A mean.

what did you learn about learning a new programme?

Language from that I have I love doing like new programme like tokers and doing the metal is for rest. IT keeps some of its upsetting J B, M. Roots, but IT is a much nicer. In fact, I almost don't know why cotton took off and not scala. I think scala has some beauty that cartland lacked where cotler felt a lot more. I mean, I was almost like I even know that actually was a response to swift, but that kind of felt like cotton looks more like swift and Scott looks more like, more like a functional program language, more like like a no canal.

Or let's just just explore. We'd touched a little bit, but just on the art, the size in the art of programing. Oh, for you personally, how much your programme .

is done with GPT currently because .

you priorities implicity so much.

Yeah, I find that a lot of IT is noise. I do use the s code um and I do like so matter of lot of complete. I do like like a very um I very like feels like girls best complete like in order complete is going to complete variable name for me so but I can just I does nice, but I don't want an auto complete. You know I hate when auto complete when I type toward for and I like puts like two two practices and two seven cones .

and two braces and like what mean without the s code in GPT, with cortex you can name, you can kind of brain storm. I I find i'm like probably the same as you, but i'd like that journal code. And you basically disagree with IT and write something simpler.

But to me, that somehow is like inspiring and makes me feel good. That also give fies the simplification process because I like, oh yeah, you don't A I system you think is the way to IT. As I have a similar thing here.

I just constantly reminds me of like like bad stuff. I think I tried the same thing with rap, right? I try the thing with rap.

And I much Better. Rapper, but like, even tried, I was like, okay, can we get some inspiration from these things? For some are lyrics. And I just found that I would go back to the most, like Cindy troops and dome schemes and i'm like, yeah, this is what the code looks like to I I think.

you know probably have different thresh hold for quenching code. You probably hate quench code that's for you and bullet played as a part of code like some of the yeah and some of IT is just like faster look .

up because .

I don't know about you but I don't remember everything like I don't i'm not loading so much of my memory about like um we have different functions, library functions on all that kind of stuff like this. The educatin is very fast at standard stuff. I like standard library stuff, basics of everybody users .

yeah I think that. I don't know. I mean, there's just a little of this in athon. And maybe if I was quoting more another languages, I would consider IT more. But I feel like python already to such a good job of removing any boy that played .

that sure is .

the closest thing you can get the suit code right. Yeah.

yes, I like. Yeah, sure.

If I like. yeah. great. GPT. Thanks for reminding me to free my variables. Unfortunately, you didn't really recognize the scope correctly and you can free that one. But like you put the freeze there and like .

I get IT five whenever I use fire for certain things that design or whatever, yeah it's always you come back. I think that's probably closer. My experience with five, you experience the programing with GPT is like you just frustrating feel worse about the whole process of design and art and what whatever I use five before. Still like I just feel like later version stupid. I am using um GPT as much as possible to just learn the dynamics of like these early versions because they feels like in the future you'll be using a more more.

It's like I don't want to be like for the same reason I gave away on my box and swish to kindle cause like I how longer would get to have paper books like thirty years from now like I wanted learn to be reading and kindle even though I don't enjoy as much and you learn to enjoy IT more in the same way I switch from, let me just pause. I switch from e max. V S code. Yeah.

I want to V S code. I think I so much.

but it's tough. And I vim to V S code is even tougher because e max is like old, like, more outdated, feels like IT. The community is more outdated.

Women, like, pretty viBrant .

still like, yeah, you wrote some stuff and lip, yeah.

but I never used any of the plugins in them. Other, I had the most one of them. I have a syntax I liter. I even complete like these things.

I feel like help you so marginally that like and now, okay, now V S code order complete has gotten good enough that like OK don't set IT up. I just got any good base and ninety percent of the time. Okay, cool. I'll take IT, right. So I don't think i'm onna have a problem at all adapting to the tools once they're good.

But like the real thing that I want is not something that like tab completes my code and gives you an idea as the real thing that I want is a very intelligent pair programmer that, uh, comes up a little pop up saying, hey, you wrote a bug online fourteen and here's what IT is yeah. Now I like that. You know what does a good job of this? Mi pie, I of my life, my pie is fancy type jacker for python um and actually I tried like microsoft t release one two and he was like sixty percent false positives. My pies like five percent false positives, ninety five percent of the time IT recognises I didn't really think about that typing and interaction correctly.

Thank you. My pie. So you like a type painting you liked, you like pushing the language there was being a typed .

language is absolutely. I, I think option al typing is great. Me, look, I think that like, it's like meat. The middle, like python has this option of typing, and like c POS plus has a .

plus allow you to take bos.

plus would have you brutally type out STD string iteration or right bottle, which is nice. And then python used just have a what type is a so um a coin sdr. 嗯， okay, so strong. cool.

Yeah.

I wish there, I wish there was a way like a simple way in python to a like turn on a mode which would enforce the types.

Yeah, I give a warning when there's no type like this.

Well, no, no. To give a warning where like my pie is a data type checker but i'm asking just for a one time time checker like there's like wait, like hattin, but I wish was just like a flag like I on three dash t oh.

I see yeah I see .

force time one type .

yeah I feel like it's a Better programme that's a kind of test, right? The type, the type remains the same.

I know that I like meat. I up but again, like my pie is getting really good and I love IT. Um and I can't wait for some of the use tools to become A I powered.

I like I want A S reading my code and giving me feedback. I don't want a eyes writing half. I start a complete stuff for me and wonder .

if you can now take GPT and give IT a code. You vote for functions, say, how can I make this simpler and have IT accomplish the same thing? I think you'll get some good ideas on some code. You maybe not code you right um for tiny grade typo code cause that requires so much design thinking but like other kinds of code.

I don't know I don't loaded the plug in maybe like two months ago, I tried IT again and found the same look, I don't doubt that these models are going to first become useful to me, then be as good as me and then surpass me. But from what i've seen today, it's like is like like someone you know occasionally taking over my keyboard that I heard from fiver ideas about .

how to pog the coder, basically get a Better debugger as IT is really interest I me.

but it's not a Better debug. I guess I will love a Better debug.

Yeah, not yet. Yeah, but he feels like it's not too far.

One, one of my coworkers says, uses them for prepayments like every time he asked to, like, just like when he needs. The only thing I can really write is like, okay, I just wants to write a thing to like print the state out right now.

Oh, that definitely is much faster, is printed. Mas, yeah, yeah. I see that myself using that a lot just is out .

functions yeah. Pro, yeah. You know what? I think like I think in two years, i'm going to start using these plugins yeah a little bit. And then in five years, i'm going to be heavily relying on some A I augmented flow. And then in ten years.

think you'll never get hundred where the like. What's the role of human that IT converges to as a programmer? Do you think it's all generate?

I ish becomes, I think that's over humans in general. It's not this programing.

Everything becomes small and small.

In fact, I was like, what the last names of humanity is gonna you? There's a great book. And its, if I recommended metaphors, prime elect last time, there is a equal called a casino autism in cyberspace. And I don't want to give away the ending of this, but IT tells you what the last remaining human currency is, and I agree .

with that will lead as a car hanger. So no more program is left ah that's .

what we're going and made code is hand and cold, doesn't have that machine polite to IT and has the sliding perfections that would only be written by person.

I wonder how far away we are from that. I mean, there's a some aspect too. You know an instagram your tittle is listed as prompt.

right? Thank you for noticing. I .

don't know if ironic or not, uh, more Sarah asc or none. What do you think of prompt engineering as a scientific and engineering discipline or maybe and maybe art form?

You know what? I started coma six years ago. I started the tiny court a month ago. So much has changed. Like i'm now thinking i'm now like I started like going to like similar coma processes to like starting your company, like i'm going getting an office and Sandy ago going to bring people here. I don't think so.

I think i'm actually gonna do remote, right? Orge you going to do remote? You hate remote yeah but i'm not onna do job and interviews the only way you going to get a job as if you contribute to get up right. And then like like like interacting through github, like like github being the real like project management software for your company. And the thing pretty much just is a github about is like showing me kind of with the future of OK.

So a lot of time is going to discord are kind grant discard and through out some random like you know can you change and set of having log in exp as alos change to log to an x to it's pretty small change I can you like change base a um that's the kind of task that I can see N A I being able to do in a few years. Like in a few years I could see myself describing that. And then within thirty seconds of power, quest is up that does IT and IT passes my C I.

And emerging, right? So I really started thinking about, like, what is the future of jobs? How many a eyes can I employ at my company? As soon as we get the first tiny box up, i'm going to stand up a sixty five vlama in the discord. And it's like, yeah, here's the tiny box is just like .

his child of this. Basically, I like you said, we need is the most human jobs will eventually be replace with with prompt engineering.

Well, prom engineering kind of, is this like, as you like, move up the stack, right? Like, okay, there used to be humans actually doing arithmetic by hand, and they used to be like big forms of people to doing doing plus and stuff, right? And then you have like press, it's right then okay, but spread, you can do the plus for me.

And then you have like macros, right? And then you have like things that basically just are spread sheet under the hood. I like like accounting soffer um as we move further up the abstraction and what's at the top of the abstraction stack or problem? Yeah right.

What is what is the last thing if you think about like humans wanting to keep control? Well, what am I really in the company? But a prompt engineer, right?

Is there a certain point where the A, I will be Better at writing proms?

Yeah, you see the problem with the A I. Writing prompts a definition that I always like to. A, I was A, I is to do what I mean, machine, right? A I is not. The the computer is so pathetic. IT does what you say so, but you want to do I mean machine ah, right? You want the machine where you say, you know, get my grandmother out of the burning house IT like reasonably takes your grandmother puts on the ground, not lift her thousand feet above the .

burn in house or fall .

but .

you know but um is not going to find the meaning, I mean, to do what I mean IT has figure stuff out. sure. And the thing is that may be asked to do is run government for me and do .

what I mean very much comes down to how a line is that I I with you. Of course, when you talk to an A I that's made by big company in the cloud, the A I fundamentally is alien to them, not to you. And that's why you have to buy a tiny box.

You make sure the a is a line to you every time that they start to pass, you know, A I regulation of G, P, U regulation on the sales of tiny boxes. Spike, it's going like guns every time they talk about gun regulation. boom.

Councils in the space of A I, your anana a bouser believer. I'm an informational anarchist.

Yes, i'm an informational and arrest and a physical. I do not think anarch in the physical world is very good because I exist in the physical world, but I think we can construct this virtual world where I can hurt you. Write, I love that title of the creator.

Uh, tweet, uh, your cyber bullying is a real man. Have you tried turning off the screen? Close your eyes like.

呀。 What had you prevent the A I, from basically replacing all human prom engineers, where there is like self, like, where nobodies the prom engineer anymore? So autonomy, great and great. And I just boys headed. Does one person's can say, run everything for me?

You see, I look at potential futures, and as long as the A S go on to create a viBrant civilization with diversity and complexity across the universe, more power to them, or die. If the A I is go on to actually, like, turn the world in the paper clipsed, and then they die out themselves, well, that's terrific. And we know what that to happen.

So this is what I mean about like robustness. I trust robust machines. The current a as a sona robust like this comes back to the idea that we've never made a machine that can self right. But when we have if the machines are truly robust and there is one prompt engineer left in the world. Hope you doing good man, hope you believe in god like, you know no go by god, go go forth and and and good universe when you .

mentioned that because i'm I talk to mark about faith and god, you said you were impressed by that. Um what's your own belief in god and how does that affect your work?

You know I never really considered as Younger. I just like my parents are the s so I I at S I never really considered how absolutely like silly ates, because like I create worlds every like game creator or like how are you an avid pro? You create world man, big bang and stuff.

Yeah I mean, what's the sky mth origin story? An sky rim. I'm sure there's like some part of IT in sky room, but it's not like if you ask the creators the big bang is in universe, right?

I'm sure they have some big bang notion in sky rim. But officially is not at all how sky room m is actually created. IT was created by program is in a room so like, you know, struck me one day how just silly atheism is right like of course we were created by .

god the most obvious thing yeah that .

this is a nice way to put .

IT like where such powerful creators ourselves um it's silly not to consider there is creators even more powful than us yeah .

and then like I also just like, I like that notion. Notion gives me a lot. I guess you can talk about what gives a lot of religious people is like, gives me comfort if, like, you know what, if we messed IT all up and we die out the same .

same way that the video game kind is confident .

IT don't try again.

Or there's baLanced, like somebody figured out a baLanced view of IT like that, like, so IT all makes sense in hand. Like a video games usually not gna have crazy, crazy stuff.

You know, people will come up with a like, a well, yeah but like man who created god, 我， that's god's problem. Like, i'm this if god, god .

living in this game.

I mean, to be fair, like god didn't believe in god.

what do you think is the greatest computer game while time do? Do you have any time to play games anymore?

Have you played the number four? I have not played the number four.

I will be doing naturally. I have to right? There is a so much as you would want to. And 对。

you don't know what i'm going to say.

where the workers?

Ft, who? And it's not that the game is so is such a great game. It's not.

It's that I remember in two thousand five, when he came out, how would open my mind to ideas? IT opened my mind to like, like this, this whole world we've created, right? There's almost been nothing like IT since. Like, you can look at him a mose today. And I think they all have lower user bases in world of warcraft, like even lines, kind of cool.

But but to think that, like everyone know, you know, people are always like to look at the apple heads at like what what do people want in this VR? Everyone knows what they want, ready, play a one and like that. So i'm going to say where the word rafts, and i'm hoping that like games can get out of this whole mobile gaming doping pump thing. I like the worlds.

The words and words are captivate a very large fraction of the human population.

yeah. And I think it'll come back, I believe.

But everyone like really, really pull you in deep's do a good job.

I mean, the other, like two other games that I think are very nowhere for me. A M, G, T, F.

sky, um, yeah, that's probably number one for me. E, T A, ah, what? What is IT about? G, T A, G T is really, I I guess G, T, A is real life. I know there's prostitutes in exit.

Real like to.

yes, I know, but it's it's how imagine your life to be actually.

I wish was that call?

Yeah yeah I guess you know because there sims, right? Which also game I like, but it's a game of head version of life. But IT also, I would love a combination of sims and G, T. A. The more freedom, more violence, more ronis, but would also like a ability to have have career and family.

And this what i'm really excited about games is like once we start getting intelligent in our, oh yeah right, like the NPC and games have never been.

but conversationally, oh, in everywhere.

E in like you like every way, like when you are actually building a world and a world in build with intelligence, 哦 yeah right and it's just hard。 Like there's just like like you run a world worker FFT, like you are limited by you where you run on A P you know how much tell only flops that you have, right? But now when i'm running a game on a hundred paid flow machine, once five people, i'm trying to make this attack. Twenty paid philps of compute is one person of computer.

I'm trying to make that a unit, one part of us, this one person, one .

person. What's all about is how much of horses, what the person .

of computer I, that's interesting. A V R, also, as a means creating worlds, you know.

But quest to, I put IT on, and I can't believe the first thing they show me is a bunch of scrolling cloud in a facebook log in screen. You have the ability to bring me into a world yeah. And what did you give me a pop up, right? Like and this is why you're not cool, mark.

That but you could be cool. Just make sure on the quest three you don't put me into clouds in a facebook log in screen. Bring me to the world. I just .

try question. He was awesome. But here that guys .

I agree with that so I was so I know because .

um I mean the beginning um what is the thought Howard said this about, uh the design of the beginning of the games decreases like the beginning is so important. I recently played zell up for the first of the breath of the while, the previous one and like it's very quickly you come out of this like a within like ten seconds, you come out of like a cave type place in a this world opens up that, and you like IT pulls you in. You forget whatever troubles I was having, whatever like I got to play .

with the beginning I played for like an hour.

the friends house. H no, the beginning they got, they did really well, the expensive ss of that space. Um the the peaceful and said that they get this the music means so much .

of that is creating the world .

and pulling right. Dw for something. I mean, there's A M mentality also, but something about VR really incredible. But the the new quest three is mixed reality, and I get to try that. So augment reality and video against done really.

really well as the other one. Is that one plus to your cameras?

I don't know. I don't know how really is .

I don't know anything you know is a .

january at some point.

some of optimism. But apple, I will buy IT I don't care if it's expensive and doesn't I will buy IT I will support this future enew .

you're the meme. Oh yes, I support competition. IT seem like quest was like the only people doing IT. And this is great that they're like.

you know what? And this is another place will give some more respect to our ochberg. The two companies that have endured through technology, apple, microsoft, I am what they make, computers in business services, all the mean social ads. They all come and go. But you want to indoor built out where yeah .

and you know does does a really interesting job. I me I maybe i'm newby this but it's a five hundred and that country and just having creatures run around the space, like our space for here to me. Okay, this is a very like balmer statement.

But IT added windows to the place the I heard about the aqua. Yeah, aqua. But in this case was something begin, whatever doesn't matter. But just like IT, IT modifies the space in a way where I can IT really feels like a window and you look out it's pretty cool. Like I was just like example game, they're running at me, whatever. But what I was enjoying, the fact that there's like a window and they're stepping an objects in the space that was a different kind of escape also because you can see the other humans, so it's integrated with the other humans is really .

and that's why it's than ever that the A I is running on those systems are aligned with you. They're gonna met your entire .

world oh yeah and that those eyes have A I mean, you think about all the dark stuff like like sexual stuff like of those eyes threatening that could be hunting like you say, like thread me in a non video game way I yeah yeah yes like they're know personal formation about me and it's like and then you lose track of what's real, what's not like.

What if stuff is a two directions the A I golf end company can take? There's like the high brow, something like her, maybe something you can talk to. And this is and then there's the lobo version of IT, where I want to set up a broffin in time square.

Yeah, it's not cheating. If it's a robot, it's A V, R. experience.

If is there in between?

No, everyone. Have you decided .

you no figured out .

to see what the technology goes?

I would love to hear your opinions for georgia's a third company, what to do a the brother times square or the the heart experience. Uh, what you think company number four be you think .

theyll be a company number, there is a to think about now. Didn't none of that tech exists? There's a lot to do in company number two. Company number two is going to be the great struggle the next six years. And of the next six years, how centralized this computer going to be, the less centralize computers going to be.

the Better for a chance we will have. So you're a bearing, you like the flag back for open source, distributed, sent, the centralization of, uh, computer.

We have to, we have to, or they will just completely dominate us. I shot a picture on stream of a man in the chicken farm. I have see that was like, factory farm chicken farms.

Why does he dump in all the chickens? Why is he smarter? He's smarter, right? Some people, some people on twitch would like he's bigger than the chickens. yeah.

And now here's a man and a cow farm, right? So what has nothing to do with their size and everything to do with their intelligence? And if one central organization has all the intelligence, you'll be the chickens and they'll be the chicken man.

But if we all have the intelligence, we're all the chickens. When are all the man? We're all the chickens.

Chicken man is no chicken man. We're just chickens in miami.

He was such a good life.

I'm sure he was. I'm sure he was. What have you learned from launching a running coming eye and tiny corp.

So this starting a company from an idea and scaling IT. And by the way, i'm all in a tiny box. I'm here. Oh, oh, I guess it's prior only now .

I want to make sure it's good. I want to make sure that like the thing that I deliver is like, not gonna be like a quest to which you buy and use twice. I made Better than a quest which you bought, used less than once statistically.

Well, this is a bit of program for a tie box amend to. Sounds good. So I will be the one I uh be the tech savi user of the tiny box just to be what the early days, what have you learned from our building these companies?

The longest time I come on, I asked why why you know whether I start a company, whether I do this? Um you know, what else was I going to down .

so you like? You like bring ideas to .

life with comma IT really started as an ego battle .

with you on.

I want to be I I like, I saw a worthy adversary. You know, here's a worthy adversary who I can beat itself driving cars. And like, I think we kept pace and I think he's kept ahead. I think that was ended up happening there. Um but I do think comma is right comes profit ball like and like when this drive GPT stuff starts working that IT, there's no more like bugs in a loss function right now we're using like a hancock stimulate the no more box this is going to be yet this is they're .

run up to driving. I hear a lot really a lot of props for pile for coma. So it's Better .

than necessity and autopilot in certain ways. IT has a lot more to do with which feel you like we lower the Price in the hardware of fall ninety nine. You know how hard IT is to ship reliable consumer electronics that go on your windshield. We're doing more then, like most sulfone companies.

have you pull that off by the way, shipping a product that goes in a car? I know .

I I N T line, it's all I may call the .

boards .

in house in santiago.

L I are shop with great testing.

Our head of open pipe is great. I like, you know, okay, I want all the cover series to be identical. Yeah and yeah I mean, you know, fta ninety nine, if dirty day money back guarantee IT will they will blow your mind at what you can .

do is the harder scale.

You know what? There's kind of downsides to scaling IT. People are always like one of you advertise.

Our mission is to solve self driving cars while the leadership of intermediaries. My mission has nothing to do with selling million boxes. torture.

Do you think IT is possible that the comment is sold only if I .

felt someone could accelerate that mission and wanted to give IT open source and like not just wanted to. I don't believe what anyone says. I believe in send us if a company wanted to buy coma with incentives, want to keep in coma doesn't stop at the cars, the cars.

That is the beginning. The device is a human head. The device has two eyes. Two years, he break air as a mouth.

So you think this goes to embody robotics.

We have we come about still. You know, they're very they're very rudimentary. But one of the problems that we're running into is that the common three has about as much intelligence as a bee. If you want a humans worth of intelligence, you're gona need a tiny rack, not even a tiny box. You gna need like a tiny rack, maybe even more.

How does that? How do you put legs on that?

You don't and there's no way you can you you can act wirelessly. So you put your tiny box or your tiny rq in your house and then you get your comment body and your comment body runs the models on that. It's it's close, right? It's not you don't want to go to some cloud, which is you know thirty million seconds away. You go to a thing which is point one mile seconds away.

So the AI girlfriend, we have like a central hub in the home. I mean, eventually.

if you fast forward twenty thirty years, the mobile chips will get good enough to run these eyes. But fundamentally, it's not even a question of putting legs on a tiny box because how are you getting one point? I have killed LED a lots of power on at that, right? So you you need there are very us logistic businesses.

I also want to build all of come as training computers. And I come to build training computers right now. We use commodity parts.

I think I can do cheaper. So to build tiny crop is going to not to sell tiny boxes. Ti box is the consumer version. But all the training .

have you talked to on.

have you talk to to work?

Would you love about Andrea party? He d to me, he's one of the truly special humans.

We got me like, you know, his dreams are just a level of quality so far beyond mine. I think I can help myself like it's just, yeah.

he's good.

He wants to teach you yeah, I want to show you that i'm smarter than you.

Yeah, he is no. I mean, thank you for the sort of honest, the raw, authentic honesty. I mean, a lot of us that I think andrey as legit as he gets in, that he just wants to teach you in a curiosity that just drives them then just like his at the stage where he is in life to be still like one of the best thinkers in the world. It's crazy like to a what is IT Michael gan micros?

Yeah, inspiration of ti. got. And that the whole, I mean, his C S two thirty one and was this was this was the inspiration. This is what I just took in run with and ended up writing this.

So you know but I mean to me that .

don't get work with dart vader, man.

I mean the the flip side to me is that the fact he's going there is a good sign for open. I I I think you know I I like alias cover a lot. I like those. Those guys are really good at what they do.

I know they are and that's kind of what's even like more. And you know what, it's not the OpenAI doesn't open forth the weight of GPT for it's that they going from a congress and that is what upsets me. You know we had two effective altruist sams going from the congress. Want to jail.

I think you on parallel on .

the you .

give me a look.

No, I think I think a factor alterations ism is a terrible .

evil ideology. Oh yes, that's just think that there's something about a thing that sounds pretty that kind of get this into trouble .

because you get can treat the sand bank ant free is the embodiment of effective alternative m utilitarian ism is in a horrent ideology like like, well, yeah, we're gona kill those three people to save a thousand of course yeah right there's no there's no underlying like there's .

just yes yeah that but to me that's a bit surprising but it's also in retrospect, not that surprising. But I haven't heard really clear kind of um like rigorous analysis why effective altruism is flawed. Oh well, I think charity is bad.

right? So what is charity but investment that you don't expect to have a return on, right?

Yeah but you can also think a charity is like is you would like to see a so allocate resources an optimal to to make a Better world and probably almost always .

that invoice starting a company yeah, right. Because more efficient if you just take the money and you spend IT on malaria. S, you know, okay, great. You made hundred malaria that. But if you teach.

no, yes, no promise. Teach hard fishing might be harder. Starting a company might be harder than out any money to already have.

I like the flip side of effective value orison effective acceleration ism. I think acceleration ism is the only thing that ever left to people out of poverty. And the fact that food is cheap, not we're giving food away because we are kind hearted people.

No, food is cheap, and that's the world you want to live in. U. B, I, what a scary idea.

What a scared of idea. All your power. Now your, your money is power. Your only source of power is granted tue by the goodwill of the government. What a scary idea.

So even think long term, even i'd rather .

die than need you be able to survive. I mean.

What a survivals basically guaranteed. What if our life becomes so good.

but you can make survival guaranteed without U. B. I.

What you have to do is make housing and food are cheap, right? like? And that's the good workout. And actually, let's go into what we should really be making, dirty ap, which is energy, did that energy that you you know that that if there is one on pretty center politically, if there is one political position I cannot stand, it's deceleration. Is people who believe we should use less energy, yeah not people who believe global warming is a problem, I agree with you, not people who believe that, no, the saving environment is good, I agree with you, but people who think we should use less energy, that energy usage is a moral bad. No, no, you are asking you are you are diminishing humanity?

Yeah, energy is flourishing. I have created flashing of the the human species.

How do we make more of that? How do we make a clean? And how do we make just, just, just how how do I pay no twenty cents for a meg water hour instead of a killer water hour?

Part of me wishes that elan one into nuclear fusion versus twitter. Part of me, or somebody.

somebody like elan have, you know, we need to wish, I wish there were more, more reels in the world. And I think elan sees IT is like this is a political battle that needed to be for at and again. Like you know, I always asked the question of whatever I disagree with him. I remind myself that he's a billionaire. I'm not so you know maybe got some .

figured out that I don't may be doesn't humility, but at the same time, me is a person who happens to know him. I find myself in that same position. Sometimes even billionaire need friends who disagree and help them grow. And that's a difficult, that's a difficult reality.

And IT must be so hard. That must be so hard to meet people what you get to that point where fame.

power, money, everybody is sucking .

up to you see, I love now, haven't shit, you know, like going to trust me. There's nothing I can give you. There's nothing worth taking for me, you know.

yeah, I takes a really special human when you have power, when you have fame, when you have money, to still think from first principles, not like all the attention you get towards you, all the admiration, all the people saying.

yes, yes, yes. The hate too.

and the hate was, the hate makes you want to go to the yes people because they hate exhaust you. And the kind of hate that you own gun from the left is pretty intense. And so that, of course, drives them right and loses baLance. And though.

and that gives this absolutely sign up, political divide alive so that the one percent can keep power, like, yeah.

I wish would be less divided, because IT is giving power to the ultra werle think the rich get richer. You have love in your life. Has love made you a Better or a worse programmer? Do you keep productivity?

metrics? No, no, no. Um I think that there comes to a point where if it's no longer this all I think I just can't enjoy IT. I so this erly love programme the minute I started like so that's .

one of the big loads of your life is programmer.

I mean just my computer in general. I mean, you know I I tell my my golf, my my first love is my computer, of course and like, you know I I sleep with my computer. It's therefore a lot of my sexual experience like come on so he's everyone is right I know you ve got to really about that and .

like not just like the I D for programme is just the entirety of the competition machine .

the fact yeah I mean, if you know I I wish I was a and some day would be smarter and maybe weird for this but I don't discriminate man. I'm onna discriminate bio stack life and still can .

stack life like so the move the computer starts to say, like I miss you I started to have some of the basics of human intimacy it's over for the moment. V, S code says, no, you.

no, but s code is no. They're just doing that, microsoft doing that, to try to get me hooked on. And I see, do IT all though, is gold digger.

Man, is gold digger. Open source?

Well, this is if it's open.

So ah I don't know.

Look, I think microsoft, again, I wouldn't count on that to be true forever. But I think right now microsoft is doing the best work in the programing world like between you get up, get up actions, V S code, the improvements to pile on IT works mysore t like .

this would have thought. Microsoft and mark zuker spearheading the open source movement.

right? right? How how things change .

is beautiful.

by the way. That's what i'd bet to replace google, by the way. Microsoft, microsoft, I in a dell. Straight up on comment .

for interesting. So your bet who wins A G I I would not .

be surprised if in the next five years being overtakes google as change .

in won't .

surprise me.

interesting. I hope some start of us.

There might be some start up to I I would equally bad .

on some start up. Yeah, i'm like fifty, fifty maybe that's naive. I believe in the power of these language models.

microsoft.

Yeah, it's great. It's great. I like all the innovation in these companies. They're not being stale into the degree they're being stale. They're losing.

So there's a huge center to do a lot of exciting work and open source work, which this. Your older, your wiser was the meaning of life. George. Hot to win .

is .

still to win.

of course.

always. Of course, what's winning was like.

no, I haven't figured out what the game is yeah but when .

I do and I was bigger than solving self driving, it's bigger, uh, the democrazy decentralizing compute.

I think the game is to stand out eye with god.

Wonder what that means for you, like at the end of your life, what that would look like.

I mean, this is what? Like, I don't know. This is some, this is some.

This only to me go trip of mind. You know, like, you understand, I eye with god is plessis man? Kay, I don't know, I know.

I don't know. Vote said god, I think you like wanted that. I mean, I certainly want them from my creations.

I want my creations to stand out eye with me. So I wouldn't, god want me to stand out eyes with him. It's best I can do golden well.

i'm just imagining the crater of video game having to a look and stand at eye with one of the characters I only watched .

season one of west world. But yeah we got to find amazing and solve IT like.

yeah, I wonder what that looks like. IT feels like a really special time in human history where that's actually possible. Like there's something about A I that's like where playing with some weird here.

something really weird, I wrote a log post. I rewrite genesis and just looked like, give you some clues at the enter genesis for finding the garden evident and i'm interested. I'm .

interest well, I hope you finds just that, George, you one of my favorite people. Thank you for doing everything you're doing in in this case, for fighting for open source, for desensitization AI. It's a fight worth fighting, fight worth winning.

Hashtag, I love you, brother. These conversations always great. Hope to talk to you many more times. Good luck with tiny corp.

Thank you. Great to be here.

Thanks for listening to conversation with George hot. To support this part, guests please check out our sponsors in the description. And now let me leave.

You are some words from Albert einstein. Everything should be made as simple as possible, but not simpler. Thank you for listening, and hope to see you next time.