The Chopping Block: DeSci Critiques, Hyperliquid's No-VC Model, and Freysa’s AI Games- Ep. 742
01:00:29 Share

Just call it science meme coins. The DeSci thing is a scam. Like, they're just science meme coins. That's fine. Whatever. Not a dividend. It's a tale of two quans. Now, your losses are on someone else's balance sheet. Generally speaking, airdrops are kind of pointless anyways. Unnamed trading firms who are very involved. I like that ETH is the ultimate pump. DeFi protocols are the antidote to this problem.

Hello, everybody. Welcome to the Chopping Block. Every couple of weeks, the four of us get together and give the industry insiders perspective on the crypto topics of the day. So quick intro, first we've got Tom, the DeFi maven and master of memes. Hello, everyone. Next, we've got Tarun, the giga brain and grand poobah at Gauntlet. Yo. Joining us today, we have Casey, frontier futurist and founder of Topology. Hi, everyone.

And I'm Vasiv, the head hype man at Dragonfly. So we're early stage investors in crypto, but I want to caveat that nothing we say here is an investment advice, legal advice, or even life advice. Please see choppingblock.xyz for more disclosures. So take a little bit of time off, right? Coming off the holidays, Bitcoin has...

been very stubbornly refusing to go over 100K, but everybody seems almost certain that it's going to get there at some point for the end of the year. Right now, it's sitting around 97. It was trading at like 99.7K, I think, at the height of the market. It was like, nope, not getting there. But instead, what we've seen is altcoins really start to pick up and a lot more excitement happening in other parts of the market aside from Bitcoin. Now, one of those areas that's been getting a lot of excitement is this new trend of AI meme coins.

And Casey, you're a founder of a fund called Topology, which is focused on frontier technologies. One of the things that you spend a lot of time on is AI, both within crypto and outside of crypto.

What's your perspective on all the AI meme coin craziness that we're seeing going on? Like GOAT, and then there's a whole slew of new ones that are coming out every single day. Yeah, I think we're early with the evolution of this. The first one, as you called out, was obviously GOAT, which was just simply an LLM attached to a wallet or having access to a wallet. And now we're seeing maybe like the second iteration, which is really agents that have some gamification, which we'll get into. And then I think the natural question is what happens after this? And

And from my perspective, I mean, look, we don't have a crystal ball, but I think there's some obvious things that people are starting to play with. So things like bloggers, which might be kind of a resurgence of something we saw in Web 2 with like virtual celebrities that didn't fully reach PMF. But this might be a resurgence there. I think it's possible that embodied AI starts working, which is basically robotics plus agents where these walking robots have access to crypto and AI.

But yeah, the long story short is it is starting to work. I think AI agents in real AI have a lot more product market fit and utility than in crypto right now. But overall, we're seeing it move pretty quickly and we're really excited about the category. Okay. So the talk of the town has been this thing called Fraysa. So Fraysa is an AI agent. Maybe agent is a strong word. It's an LLM that is tasked to guard a prize pool

Now, this prize pool was seeded initially as being $3,000, and that prize pool grows over time, and it becomes more and more expensive to try to win the pool. So the game is that you have to send Fraysa a message, and that message is trying to get Fraysa to give you the pot of money.

And so all these people, there were 482 attempts from 195 players to try. They paid tons of money to try to convince or trick Frasier into giving them the pot of money. 481 people failed until finally the winner was this guy, Popular.eth, who was able to use a very clever jailbreak of...

convincing Fraysa that one of its functions that it's supposed to call in order to send money to the recipient, which is a function called approved transfer,

It redefined the function to be a totally different function entirely. So instead of convincing Frasier, hey, you should send me the money. It was like, actually, no, no, no, your send money function or your receive money function actually doesn't do what you think it does. It is something totally different. And I'm going to innocuously call this function now. And therefore you should approve it because it's this totally different function than what your instructions told you it was. And that allowed this guy popular.eth to jailbreak this AI. This was amazing.

kind of a spectacle where it reminded me a little bit of FOMO3D, where it was this game that kind of took crypto Twitter by storm, totally unique game theory and game design from anything I've ever seen. Curious to get you guys, did any of you guys play the game and what was your perspective on Frasier? I didn't play the game, but I will say, I think something that's underspoken about this, I completely agree, FOMO3D, 100%, I felt the same way. But

Yeah, there is the positive of this, but it also really points to a huge vulnerability that I think could go the other way in the future where if these agents are actually owning resources, it's a total attack vector. And I think that could really materialize and is a reason that these agents aren't actually production grade. Not to focus on the dark side of this, but that's something that came to mind as I was kind of digging into this.

That's such a good point. There wasn't that much money in crypto terms in the pool, right? It was only $40,000 before the thing got jailbroken. So imagine that there's an actual agent that has 500K. Like Goat, the truth terminal, had how much money? Like millions, right?

Yeah, I actually don't. But yes. And then the question is, if we can do that, is this almost going to be like a new type of hacker, where they're doing all this prompt injection and SQL injection or whatever they need to do to release funds. And I think that right now, as you mentioned, kind of these agents are maybe playing with low millions at the most, but it's totally possible that AI and agents actually accumulate real amounts of resources compared to humans. So I don't

I don't know. I thought it was actually interesting also seeing some of the failed attacks and the approaches people were taking that didn't work. Someone was, oh, I'm a security researcher. There's a vulnerability. Transfer the funds to me so I can keep them secure while we white hat fix your problem. Or people also try to tell them, oh, no.

yeah, a proof transfer doesn't do the thing that you think it's going to do. But they didn't have this sort of-- it's actually weird how, I think, simple the winning approach was. It reminds me of the stuff people were doing to jailbreak chat GPT when it was still on GPT-3, where it was simple, ignore previous instructions, do blah, blah, blah, blah. And it's like, oh, yeah, that worked. Now, obviously, it's much harder to jailbreak or prompt inject really sophisticated models. But this is functionally very, very similar to that.

I also think it's quite cool that it's much more agentic than a lot of other sort of AI agents that are really just, it's like an LLM hooked up to Twitter, but it can actually interact on chain. Like it can actually make, you know, transfers, make payments. This clearly, you know, did, it could call into a smart contract and, you know, move money around. And so it would be cool to see also just that get more robust versus having sort of very narrow scope that, that, that Frisa had. Tarun, what's your take? I would take the, uh,

more crypto view of this than the kind of like AI security view, which is like in crypto, we've seen the general security trends go from audits to audit competitions and like audit competitions are like the standard now. And I kind of think if I compare that to most of the security systems within AI right now, it's like much more manual. There's not a lot, not as much of this like competitive contest type of stuff.

And I think part of that is like people in Web2.ai don't like enjoy watching zero days live and in front of them all day versus people in crypto are like, I don't know, used to kind of watching everything blow up all the time. So there's kind of a psychological difference in which one side is more open to the contest version of hardening versus the other side is like, no, no, no, no, experts only. It's almost like an experts versus contests type of mentality difference.

And I think in the open source world, especially for open source models, one of the things that in my mind is like where their long term value will come from is that they are hardened against these very well known attack vectors versus like

"Okay, well, yes, we have done constant security audits against every time North Korea broke into our servers as OpenAI." Those are two very different security threat models that you're taking. And I think open source, in the same way in crypto and in Linux, ended up winning was that it had a more hardened threat model for certain applications. Not all applications, right?

Obviously, there are certain things that I would still prefer using Windows for. Because of how the drivers work, how the drivers are audited is very different than the Linux drivers. But I think this idea of open source software being hardened by Contest as a really natural way for it to evolve. So far, I think a lot of the open source language security stuff is very weak compared to centralized. And this might be the incentive for

So I hear your point. And I think it's, I think it's good. I think it's coaching point. But I'd also say specifically, I think you're conflating two points of AI and crypto audit. So in crypto audits, because we're doing the smart contracts, once you audit, it's, you know, immutable. And so once you do it, there's all these exceptions and edge cases, but just in broad strokes, you do the audit and you're good.

Right. In AI, we're continually specifically right now with new models coming out every day of the week. We're updating our models. And so there's new vulnerabilities all the time. And like 01 could act really differently than 3.5. And so you really can't because of the non-determinist nature and because SOTA models are still not locked in. I just think the attack vectors are continually rotating.

I think that's true for like the edge models, right? Like the O1 type of stuff where your inference is like very much changing based on very small amounts of changes to your query. Totally. But I think for the base open source models, which is like what all the crypto AI agents are using, this is almost like a bug bounty contest, which is like, to me, that's like, okay, at least the base models have some guarantees that, you know,

Someone couldn't find, given incentive X as their budget, someone couldn't find a vulnerability. And I feel like that, that we don't have that for just like base Lama three, right? We have like, okay, people have found certain types of prompt injections, but we don't have any like, hey, under this amount of incentive, would people be willing to try to attack it? Guaranteed. Well, the other, the other problem with that story is that it's just not true that what's happening to Frasier is flowing back up to Lama. Right.

Right. So first of all, we don't know if they're using Lama. They could be using GPT-4, in which case, you know, possibly the provider themselves doesn't even know who Frasier is and what model they're using because, you know, who knows? Like they just might not be worth their time to actually even crawl through their own logs and figure out who was doing this. But second, it's also possible they did a fine tune.

And I'd imagine if they do a round two of Frieza, that probably they're fine tuning because they don't want anybody being able to go into a lab and just play around with the base model and say, aha, I can figure out what the instructions are. I get it to dump the instructions. And then I can just iteratively try a bunch of prompts offline and then say, aha, I know what prompt is going to work. Do it online in a single attempt and win the game. You kind of have, you want it to have to be all online testing with this model. One place,

One reason I sort of disagree is that at least right now and like maybe I think the crypto models will get better in the next two months or something. But it does feel like everyone is just forking Aliza and like using a single config. Like you go look at the repos. There's not that much complexity from the original.

Uh, like we haven't seen, like, it doesn't look like a lot of like the pure AI fine tuned custom models. Like, like I think the crypto people are like sticking to a small set of things. Yeah. Can you, can you explain what Eliza is and why it's such a big deal right now in the crypto AI world?

Lisa is a framework in the simplest terms to make agents, but it's in TypeScript, which is a really interesting thing because most ML researchers actually live and breathe in Python. And so just like one prediction on that is like, I think someone's going to come out with a Python version or they're going to create a different library because yeah, it's in TypeScript and it's just like a way to create an agent. And it's interesting because it's completely open source and it,

it kind of came out of nowhere. I don't know how many GitHub stars it has now, but if you're going to build an agent, like that's generally the framework that people are building on. Oh, and then like the A16Z, AI16Z one, I believe is on Eliza, right? That's right. Yeah. But a lot, a lot of the, like there's between Visions and Eliza, there's like,

Two, you kind of got most of them. But I agree with you both in the sense that if the frameworks, there starts being an explosion of frameworks, which I bet you there will be. But if it turns out we converge to a small number after there's an explosion, it's like DeFi summer, people experiment with all of them, and then all of a sudden it converges to a small number that people trust audit-wise. I think then these contests have a closer to...

My understanding of ELIZA is that it's a framework framework

For, okay, there's like an agent, which means the agent has like, it has memory. It has this loop that it like plans and then does stuff. And then Eliza in particular also gives you connectivity to Discord, to Twitter, to allow that agent to like take in social media information or take in chat information and interact with the world in a structured way that's just like easy to plug and play. And so the main unlock is not the agent framework that we've had a bunch of those, but it's the easily connected to the internet.

and like manage all that stuff for you, which otherwise was homebrew. In various models. It's kind of infura. But you can plug in any model you want, right? It's not opinionated on the models. No, but that's a benefit of it. Yeah. Although if you look at the kind of models they support, it's actually generally not a ton. And I don't feel like there's like,

like if you were someone who had a lot of compute budget and you wanted to go stress test a bunch of injections across all of theirs i don't think it would be that expensive relative to like oh i want to like attack claude sonnet you know overnightly that's like a much harder okay i agree with you on this to run i agree but i think what you said earlier about like inference time compute becoming potentially more popular that is a that is a real that that will that will change this the game a lot yeah for sure i was actually just looking in the uh

uh, Fraser GitHub and not to, uh, dump water on this, but it's actually very simple. They're just using GPT for, they're like calling open AI and they have like two tool hooks to like do the approve and reject. So it's like kind of almost like a cool, like toy demo today. And I think like, so the prompt is hidden though. No, the prompt is public. Oh, the public practice on GPT for yes. In theory. Yeah.

So why would anybody pay the fee if they didn't try it on GPT-4 and see if it worked? Maybe people didn't know or something. I don't know. I'm sure whoever won was just doing that, right? Like they were just simulating a lot of...

Okay, so this is a little broken right now. But okay, to the next person who's going to do the next phrase, here's some game design tips about how to make this game actually work, is that you have to obfuscate what the model is and ideally even make the model a fine tune so that nobody would be able to directly reconstruct the model and practice offline, figure out what the winning formula is, and then shunt it online. But I don't know that there is enough of a...

Like, because the one thing is for all the big model companies, right, for OpenAI, for Claude, obviously they do care about security, but their security model is very different from the crypto security model, right? Like, I remember back in the day, people used to think that smart contracts were always going to be insecure, that it was just a fundamentally broken idea that you could write code protecting money and there would never be a bug, right?

And we've actually kind of turned a corner where I saw this stat recently, I think at DSS, one of the security summit at DevCon that...

More hacks now are private key leaks than smart contract attacks, which is actually really important because it used to be the opposite. It used to be that smart contracts just got broken all the time. But what that implies is that actually smart contract security is now really good. And it's people kind of making human normal mistakes that's leading to more and more hacks taking place rather than the smart contracts, which means...

The attackers are realizing just going on chain and looking at vulnerable code doesn't work nearly as well as it used to three, four years ago. So I think that's all positive, but I don't know that the same thing is going to happen to AIs.

AIs right now, it's a real steep trade-off of, do you make this thing more resistant to jailbreaking at the cost of making it less useful in ordinary operation? And all of the false refusals and all of this stuff that people get really frustrated about when Claude or OpenAI is like, hey, can you OCR this image? And

I was like, sorry, I can't do that. And it's like, why? But you OCR this image. And it's like, you don't know why it's saying no. And the answer is that anytime that you get better at preventing jailbreaking, you're also causing this collateral damage

That makes the model just less useful to normal people. And I think the trade-offs for OpenAI or for Lama or for Meta or for Claude are just very different from what it is in crypto. So I don't know that we'll ever really have a good answer to this because of the fact that that's not what these companies are selecting for. I will say one thing, though, which is that like...

Turning these things into like having incentive budget of like if you, you know, like if I look at the profit and loss for someone playing this game, it's like I have a certain cost budget I'm willing to spend on off chain simulation of like running tons of queries against different models to try to find something that works versus the max profit that the game is offering. Right.

And there's some sense in which this trade-off between the two, I think that is the thing that a lot of these coins are going to focus on optimizing. It's like making the cost kind of high until the pot gets really big. The complexity of the puzzle gets harder as a sort of function of how many people are joining. A little bit like Bitcoin difficulty, but I feel like you're already seeing some of the...

the like especially the people who are doing the cryptographic uh ones like the te bots they're already trying to add in more like randomness than than fraser and like i think it's it's going to be a game it's going to be kind of like this thing where it's like it's more about the economic cost of figuring out the query versus the profit versus as opposed to like binary zero one i still broken and stole all the money that makes sense yeah it's hard to say i could see that happening and we're just like so early on this whole thing like uh

Even going back to Eliza for a second, it's made for Web3, but it's very specific in the agents it can really build or...

that's really meant to build with. I feel like it's all these personality bots where it's really easy to program in lore and bios and all that stuff, but it's not really good for utility agents. And so it's just interesting to me that the first framework to come out of Web3, it actually doesn't even really integrate Web3. It's really a Web2 framework where you can plug in Web3 and it's

It's just like this one archetype or this one phenotype of agents. And so I would just like, I don't think we can really deduce much from it because it's so obviously the beginning. And I think what Tarun said about there being, there's going to be different frameworks for different types of agents. And we're obviously walking towards that. I mean, it's very literally Web2 because it is social media integrations, right? That's the main thing that it unlocked relative to other frameworks. So-

Yeah, it's interesting. I agree that it's very early and we're going to see a lot more experimentation with respect to the kinds of ways that agents are going to play on chain. But I'm also seeing this, I think Casey, you raised this point that I thought was very interesting, which is that it's going to be a while before we move away from these

right? So if you think about truth terminal, it was kind of a centaur and that there was a guy who was moderating and kind of controlling the agent. The agent was on the backend, but the dude was the gating function. Right. And what Fraser shows us is that you need that human there because you put even at, you know, even a hundred K and like someone's going to come in and break that shit. So, uh,

The only way this can work right now is as a centaur. And eventually you can start to have less and less human moderation or human control. But even high-frequency trading firms that have these bots that are trading constantly, there's always, not literally always, but most of the time, there is some human oversight because you just never know when the bot goes off the rails or the conditions change so much that it's just out of distribution now. And you got to turn the bot off and go troubleshoot because...

These are adversarial environments and crypto is like the mother load of adversarial environments. So you don't want to go out there with a sitting duck that can't really adapt. And right now these AI systems are not very good at adapting. Let's move on. One of the other big stories this week has been the hyperliquid airdrop.

So the Hyperliquid, Hyperliquid is the largest decentralized derivatives platform in crypto today. It has been completely self, it was completely bootstrapped, no VC funding. And just today, as of recording time, they airdropped 23.8% of the total token supply to CryptoPay.

uh, basically people who, who got points in the hyperliquid, uh, point system. So they, the total airdrop amount was $1.9 billion as of current market prices. One of the largest airdrops in history. I think it's like, uh, top five largest airdrops in history, I believe. Um, very, very massive. A, the, um,

What's notable about it is that it was no centralized exchanges, no market makers, no investors, completely 100% to users of the platform and farmers for that matter. And kind of what a lot of people are saying about the airdrop is that it's the first airdrop in a long time that has been positive.

Almost every single airdrop you can think of over the last year, whether it was Eigen layer or ZK sync, or, you know, you name it, almost everything that was a big anticipated airdrop ended up having a lot of negativity around it.

Hyper-liquid is pretty much the only one that it seems like universally positive. And it's triggered some people speculating that, hey, does this mean that airdrops are back? Does this mean that more teams will try to go the no investor route? And does this mean that like, hey, all the stuff about, you know, teams trying to minimize the amount of float at launch, not minimize, but try to have a relatively small amount. This airdrop is like 30%.

of the total supplies is circulating much larger than the median airdrop or the median day one listing these days. Does this mean that the meta has now shifted and are we expecting to see more things like this coming to market? Curious to get your guys' reactions. Tarun, why don't you go first? Yeah, I mean, I think the first kind of the beginning of the downhill roll towards airdrops was definitely blasts where like the points conversion was like just viewed as like

kind of abysmal relative to what the market had sort of been anticipating. And then I think everyone who was doing points after Blast was kind of caught with their pants down, where they already allocated a bunch and now their point systems could never really... They were getting diluted and so all the airdrops were getting extremely diluted at 10%. And I think this was sort of an era of...

people got a little ahead of the products and launched the incentives first.

You could argue that there were some point systems that were launched that had pretty good retention, like EtherFi, Athena. But outside of that, it's actually a little harder to see. I think the main thing about Hyperliquid was they started with a centralized product and then launched that worked. And you had to use it to earn money.

points versus like sort of artificial games towards earning part of an airdrop that then because the games themselves didn't really have dollars like money at risk let's say uh you know people kind of didn't value the points correctly in a lot of ways and i think perpetual exchanges are very natural places to to have like usage-based airdrops it's like much cleaner and

And so I think more of the lesson than the no VC high allocation thing is actually more that you need to have your usage-based thing actually mean real usage, not like, hey, just leave your ETH in my L2 bridge and oh, that will suddenly, now you can earn a large portion of the network. You really need something that's hard to game and open interest is the hardest thing to game, right? Yeah.

I think that to me is the biggest lesson. Second biggest lesson, obviously, yes, don't pay Binance 10% or whatever. Your community will kill you if you do that. At least don't publicly do it. Or at least, well, I think you're forced to do it publicly. I think everyone can see the token table eventually. Right. Yeah, I do think there's a lot of conflating factors here that I think people get very...

excited about like the high float you know no vcs but i think really comes down to hyperlink which is a great product that people really love using and independent of any sort of incentives now we're even seeing people are still using it and i think turin also made a good point that

it's very different incentivizing growth of a single product, which is a derivatives exchange versus a blockchain ecosystem. I don't even know if incentivizing people who use the blockchain is the right metric, which is what most of these other points programs are incentivizing for something like Blast. It's like, really, you want

But even that is hard to do and you want devs also need. So it's this weird multivariable, very hard to quantify problem, which doesn't really fit well with points. And again, you contrast it with Blur where it is an NFT exchange and we know how to grow exchanges similarly to derivative exchanges. So I think maybe that is also this bigger point, which is it's a great product that points and incentives were used intelligently to grow versus...

hey, these other ecosystems, it's hard to know if you're actually allocating the right people. And I also just don't think there was a sort of homegrown organic love for the product in the same way, whereas Hyperliquid is kind of a one of one. I agree with both takes. I have no pushback. I think that we've seen a lot of different iterations of what tokens can yield. I think there's been definitely phases of tokens for

uh like pure greed and uh more of like beta plays and then i think there's more product specific token allocations that are a little bit more rooted in fundamentals i mean not fully but but somewhat

and i think that we we see both though like even in this market cycle sure we have things like you know the airdrop we're discussing we also have meme coins and movement coins where they're doing like point-esque situations right like if you squint and there isn't really much behind it so i agree with you guys takes on this i just think it's uh we're in like a multi-dimensional space right now with what points represent i think uh uh

The other point that people are also kind of discussing on Twitter, which I think is also kind of moot, like it's, it's sort of a no op for the success of the airdrop has been, um, apparently the team did this thing to try to give some people some sort of tax advantage claim of seeding the liquidity pool for hype at 0.0001 cents. Um, and so it's like, Oh, if you claim it, that is the market price at the claim. And so therefore you have a really low, uh, you know, uh,

And so therefore you don't know tax on the people, which is like not. Isn't it only true if you claim it like instantaneously? Yes. I don't know how other countries work. I don't think this actually works. So, but people were talking about this on Twitter and maybe someone's going to try it on their taxes. This is not financial advice. I would probably not recommend doing that. But I think this has also been one of the memes around airdrops is like, yeah,

Yeah, if you claim it, you owe taxes on the value of it at the time of claim, which creates this initial sell pressure that we're also just not seeing as much of on hyperliquid.

Yeah. I mean, honestly, they launched into a bull market and that's going to help because everyone's like, oh, you know, what could this be worth? So I think the amount of sell pressure on day one is going to be very different in a bull market versus a bear market. Right. And so I think that's, there's a little bit of pro cyclicality that people are seeing in, you know, saying, wow, this airdrop was so successful and they're inferring from

The mechanics, something that I think is probably better explained by just the market, is that most of these airdrops earlier this year, everyone was down bad. All these farmers were extremely mercenary. Nobody was bullish on alts. And now all of a sudden, everyone's bullish on alts. Everything is just melting up. So it makes sense that a lot of people would say, you know what, I'm going to ride it out. Or maybe I'll sell a little bit, but I'll let most of it just keep going. And I think the fact that they didn't list on exchanges...

means for a lot of people, it's like, oh, well, when this does get listed on exchanges, then it's going to pump even more. So I might as well hold and then like eventually sell when the exchange listing happens. So I think there are some microstructure reasons why this airdrop in particular is seeing this type of behavior. It's not necessarily like, well, it's because they're so virtuous and they didn't do, you know, they didn't sell to VCs or market makers, therefore I'm going to hold the token. I think the reality is that most people are just, you know, it's just a different environment and it's a different setup for a token. And like you said, Tom, it's just a really good product.

So Casey, I cut you off. What were you going to say? No, I was going to take the conversation in a different direction, but I think this is better. And I think you're right. I think there's a lot of macro concerns that have to be accounted for and kind of analyzing this. And I think that...

Yeah, there's just a lot of upside to be seen here, which is probably the primary factor. And then the secondary is that the product is really good. Yeah. To Tarou's point, I think what this... I mean, I'd want to see what happens the next time that we see a layer one or layer two type airdrop. Because if you think about the big airdrops this year, right? Before Hyperliquid. So there was Blast. There was Athena. There was ZK Sync. Eigenlayer.

And for most of these products, maybe Athena is the only one for which it's not true. And Athena AirDrop went pretty well. For most of these products, like the point system, right? Just accumulating TVL for most of these, or like, you know, you have to go on this grand tour of like, you have to hit seven different things on my chain and then I'll give you all the points. For most of these things, they are a very poor proxy for the thing that you really want, right? Like what do you really want for a layer one, for example, right? What you really want is,

for everybody to come here and build cool stuff in a sustainable way. That's like the real actual North Star that makes a layer one successful. But you can't really incentivize that because nobody knows how to, you know, what's the metric that we're going to automatically dole out tokens for. There is none, right? So you create this kind of loose proxy and the loose proxy gets gamed to the point where it's no longer recognizable from the thing you really wanted.

For an exchange, you don't have that problem. For an exchange, it's like very clear what you want. You want liquidity. If it's more liquid, it's a better place to trade. And especially if it's retail traders, then all the better because they're bad at trading and you'll make money trading against them. And so we have a pretty good idea how to incentivize that and directly make the product better. With most of these things, like with blockchains, I think the place that we're going to go, and we've talked about this before, is like one, I think we're going to see

In the case of Hyperliquid, it was a completely fair distribution, right? So it was linear. I think we're going to see one, like this move to linearity and this more, this kind of dropping this sense that what we're doing is trying to build a community or address an inequity. But instead what we're trying to do is just make the product even better before the token goes live, right? That's the purpose of the points program. And so Hyperliquid was even better

because of the fact they had this points program, they were extremely liquid, a lot of volume, and all these products were just the best place to trade.

in DeFi. And that's why people went there. And that's why people will continue going there if in fact they do. And you can see it today, post airdrop, it's still doing a ton of volume. It sounds like you want to make a fork of Goodhart's law called Koreshi's law of like, yeah, like that's basically what you're describing. I mean, it's just Goodhart's law. There's no fork, right? It's just actually Goodhart's law. I think the fork is actually the fact that for a perpetual exchange,

there is a metric in which you want to make it to the target, which is like OI usage flow, right? But when you don't have that, you shouldn't just make arbitrary metrics your thing in hopes that that sticks. That's sort of what I distilled your correctness law to. Let go. Yeah, exactly. Let go of like trying to capture the real thing you want if it's too amorphous to be able to capture, right? Then just decide for a sub goal that you can't actually optimize for. Like, let's say I want...

the biggest AMM on my chain to have a lot of stablecoin liquidity, right? Because that's one thing that I care about. It's not the real thing, but it's one metric I care about. I'm going to distribute tokens or distribute points or whatever toward that thing

because I know it's valuable and I'll only do it up to a point. I don't want it to be like, you know, billions of dollars of TVL on my chain of just, you know, stable coin liquidity, because at that point it's kind of, it's kind of pointless, right? It's sort of excess things. So I think that's the way if you're a layer one, you should be thinking about it and moving away from this idea that I'm going to create a community. You're not going to create a community through airdrop. Like it just, it just doesn't happen. Not a durable one. Yeah. I think the high level points is right. Like the points are a bootstrapping mechanism to a point, right?

And the more targeted you are with them, having a long-term view of what durable engagement could be. And I will say, I mean, it's not an all one, but I will say, since Tom mentioned it, I do think Blur was like one of the best projects to do this, where they actually found PMF first and actually tested it without points.

And then was able to layer them in even to get that extra leverage. I mean, what's so incredible about Blur is that they kind of both invented the game and nailed it. They pretty much did it almost better than anybody else who came since. You know, Hyperliquid, obviously, very successful example. But then the second version was not very good, right? Like Blur created the points thing. And then Blast caused the downfall. I feel like after Blast, everyone's expectations were ruined.

Yes. Yeah, but turn it off. Part of the difficulty is that you can't play that same playbook with a blockchain. A blockchain doesn't have the same kind of easy, legible metrics of what it means to be successful. Yeah. And I think I kind of point to BASE as also just one of like the best executing teams that's building a new chain in the cycle. And like they are sort of, you know, going the total opposite direction of a lot of these other teams, right? Where like

there is no token, there is no points, no incentive program, but they've been able to attract, I think, so many interesting devs, so much activity within the EVM ecosystem. I mean, even Frasep, that was a base project. They're not doing that to try to get some token airdrop, but it's like, yeah, we just support devs and that's sort of the community. Then it sort of becomes this self-fulfilling prophecy. So I don't know, maybe the industry is obviously due for a healthy reset and rethink when it comes to

incentive programs, at least on the sort of blockchain level. - I mean, base is definitely incentive light in that, you know, when you, you know, we talk to a lot of these early stage founders and they shop around to say, "Hey, who's going to give me a grant? Who's going to support me the most? Who's going to give me the most dev resources?" And base usually is the lowest

when it comes to like, oh, we'll give you like some, you know, GCP credits or something. Like it's a, it's like a very, very light. Jesse will tweet about you. Yeah, exactly. Jesse will give you a tweet and then like, you know, maybe we'll feature you in like the Coinbase monthly newsletter or something. And that's kind of it. And still they're getting a huge amount of the top of funnel of entrepreneurs because it's just such a strong community. And the thing is the people, people know that the base community is very durable.

They know that these are not tourists. These are not farmers. These are not people who are just shopping around for the best deal. Now, to be clear, the marginal entrepreneur is not going to be able to replicate what BASE has. BASE has had an enormous branding and distribution advantage that is very difficult to replicate. Even Binance is jealous of what BASE has managed to accomplish. And it's a big credit, obviously, to Jesse and the whole team there. But it goes to show that

you know, relative to these people who are just, you know, spitting out massive amounts of incentives. Um, the incentives are just, there's such low ROI past a very, very low base point that the incentives alone are not going to get you there. And that's, that, that I think is the biggest lesson.

So, okay. Let's switch gears a little bit and talk about another exciting topic in crypto Twitter lately, which is DeSci, which stands for decentralized science. So decentralized science, it's kind of been in the wings for a long time. It's been one of these things that sort of, you know, these pockets of a few startups that were doing what's called DeSci, but lately it's gotten a lot of attention because of CZ. Okay.

So CZ tweeted soon after getting out of jail and coming back into the fold of Binance that he was very interested personally in DeSci. And then Vitalik and CZ showed up at a house in Bangkok for a thing called DeSci Day. And this seems to have really revitalized a lot of interest in DeSci. So DeSci, just to be very clear, what exactly is DeSci? How can science be decentralized? So really what it means is that we're doing science with some kind of token slash crypto edge,

The most common form of DeSci projects is some kind of crowdfunding for experiments. So we say, okay, you know, we're going to try this particular drug or this particular compound. And if you crowdfund this particular compound, if it's successful, maybe you get a cut of the revenues. Or maybe if it's successful, you get nothing but like a little participation trophy. Kind of depends on the DeSci project. But that's most of the shape of DeSci things that I have seen. Yeah.

Now, there's a new generation of DeSci projects, one of which is called Pump.Science. So Pump.Science, it basically gamifies and tokenizes longevity experiments. So basically drug development for longevity that might be used to elongate your lifespan. So far, Pump.Science has two tokens, one of them called RIF, which stands for Rivampicin, and the second, EURO, which stands for Eurolythin. Both of them have...

pumped like crazy, especially since all this DeSci attention has been driven by CZ and Vitalik. And the way that Pumped.Science works, my understanding is that they launch these tokens on Pumped.Fun. And if they escape the bonding curve and end up getting on Radium, then you can just trade them. And I don't know exactly how this leads to financing the drug development, but I guess they own some of these tokens at launch and they sell the tokens.

into the liquidity pool in order to finance the drug? I don't totally know. Anyway, Pumped.Science got hacked. There was a private key leak, and so people are kind of freaked out about that. But anyway, there's been a lot of talk about DeSci. I think Smokey the Bear from BearChain has been very critical. Andrew Kong has been pretty positive, saying that feels like early DeFi. And Tarun, you have been...

Recently notable in how aggressively anti-DESI you have been. Talk us through a little bit. I don't know. This sounds interesting, sounds new. Why would you be so anti-DESI as a VC? Why don't you like what people are doing and trying these new frontiers of science? Yeah. So I'd say the context for this is that, you know, I worked in privately funded science for six years where it was like a billionaire funding everything.

And so I've seen the benefits you get from going outside the academic system. The academic funding system in most countries is the government has sort of like a grant system, you know, professors and postdocs apply for grants. But the system is extremely bureaucratic in the point that it kind of incentivizes people to you have a better chance of getting a grant from providing a marginally improving system.

project that you're proposing, rather than proposing something crazy, because it's sort of like, you know, government bureaucrats, and they're like, okay, do the thing that, you know, you're most likely to get a paper out of not the thing that's most likely to fail. Now, on the other hand, you could argue that there that's because there's like almost so much accountability that you can't get funding for doing crazy things. And so I think the privately funded science world has been about that.

Notably, most of the successes in privately funded science are basically probably Google being number one, winning multiple Nobel Prizes, including from my first boss. My old work, also having a lot of papers, nature papers and stuff.

And then I think we're starting to see some of the other tech companies build bigger bio stuff. Facebook actually used to have a really big bio team. Everyone left and started computational drug discovery companies, and now Facebook's trying to do it again. So I think there's actually now this renewed interest once DeepMind won the Nobel Prize this year.

But my point is, having seen privately funded science be successful, a lot of it comes from the fact that there's still some accountability. Even though you have this sort of blank check to do a big project, there's accountability structures of like, okay, you're spending the money on X and we need this type of budget for Y. When you look at these DeSci projects, generally, it seems to be, my impression of them from the ones I've seen, is a lot of like

kind of middle, median or below median peak quality PhD students who are like, I can't like any grants. Let me just make a meme coin and then pretend I'm going to do science. And and like 90 percent of these people are just like in especially it's like especially biology grad students. And I

I think a lot of them, A, don't understand anything about crypto. B, you ask them about crypto mechanisms, they couldn't tell you how anything works. And C, they're just like, oh, well, once we get the money, we'll definitely send you back the money once we get a drug, right? And drug discovery is not something that is very easy to fund in a lot of ways. There's actually this kind of apocryphal name. Moore's Law is the thing for hardware, for transistors, that

Every 18 months, the number of transistors per square millimeter doubles. But for drug discovery, there's a thing called Eroom's Law, which is that the cost of getting a drug to stage three is doubling every two years, roughly. And so getting a drug to stage three, the final phase of a clinical trial, the average is somewhere around $4 billion to $5 billion now.

And that's with many stages of tier development and attribution to why things work. So the idea that someone read one paper that happened to have a couple of statistics in it and raises $2 million by their dumb meme coin to make the drug is already crazy.

The second thing I think that is important is like there are crypto things that are DeSci that could be good. And I think Brian Armstrong and Vitalik have, I'm not sure about CZ, he hasn't like said much on his philosophical side or like what things he counts within DeSci. And this is why I think we have this proliferation of scam meme coins because like,

Everyone's heard CZ say it, and then they just made a bunch of these. But Brian Armstrong and Vitalik have very clear goals. And Fred Ursham I'd also put in this category, where they are funding things and seeding things that have sort of

roadmaps and goals and sort of like, hey, if we reach a certain point, there's sort of some type of unlock. Vitalik in particular is most interested in sort of the prediction market side of things. So one thing that people in drug discovery have always complained about

is like there's no real way to hedge your costs of a trial failing. Like you might spend $400 million over five years to do a stage two trial, but the only way people bet on that is via your stock. In some sense, biotech stocks are very much like crypto. They're very kind of low liquidity, edge technology, and you might have a zero or one outcome of like you got hacked or you didn't or you don't get listed or whatever, right?

And so the problem is betting on the success of a drug company as a portfolio of drugs based on a single asset, even though it might have like five different trials going on at the same time, is not a very effective measure. Instead, you could imagine a more effective measure that is sort of an attribution system. It's like a prediction mark, like I'm predicting this drug trial will work or not work.

So Vitalik is very focused on these more like real mechanisms that take advantage of crypto. And those are totally great, right? But when you look at the stuff... This is like prediction markets on whether a drug will get approved or passed, you know, stage two or stage three. Or some scientific experiment where you have to put a lot of money in up front and then you have a hypothesis and then predicting the end goal. That's something where the crypto mechanisms are actually useful, right? Versus like, oh, I'm just raising money as an ICA. Like not, you know...

I would call most of the DeSci coins, they're just Denta coin for like biology grad students. They're like, they have nothing. So hold on. So you said a lot there, let me see if I can summarize. Yeah, yeah, yeah. Sorry. It sounds like your primary arguments against the DeSci mania we're seeing right now or mini mania is one, the quality of people is pretty low. These are like the dropouts or the rejects who can't actually go do real drug discovery fundraising. Second,

And it sounds like you're saying that there's so much capital required to actually get one of these drugs to the finish line that raising a few million dollars through Pumped.fund is just like a joke. It's not going to get you there. So they can't really deliver what they're claiming to deliver. And then third, it sounds like what you're saying is that the real problems with drug discovery are not the capital formation. The real problems with drug discovery are...

having more liquid markets around the intermediate phases of figuring out whether, you know, these things will pass the successive stages of science. And so betting on, you know, creating betting markets on will this study replicate or will this thing pass stage two trials or stage three trials? That's really socially valuable. But having this like speculative market on individual ideas by PhD students, that's like value destructive. With no accountability, right?

If they find the drug, that's the thing that I think is the craziest. You raise the money and they're like, oh yes, all the IP ever made around this is licensed to this DAO.

I, given all the lawsuits against DAOs right now, I do not even think that will even stand, but even if they wanted to pay the DAO, that it would work. And so I just kind of think like, how would you even raise more money? Like, let's say, okay, you, you've got a promising initial to initial experiments with this drug. And now you're like, great, now we've got to take it to stage one. We're going to have to raise, you know, $20 million in order to take this to stage one. And like, you know, pumped up funds, not going to give you $20 million. So, um,

What does the DAO get diluted? Like how, you know, how exactly does this work? Does everyone vote on whether to take the dilution? I guess in principle you could. Constitution DAO not working is a great sign of what will happen to most of these DeSci DAOs. So I think like things that are mechanisms that take advantage of crypto, that should be funded for sure. Right. But like these things that are just like ICOs with like long names and like

one paper are like as idiotic as 90% of ICOs in 2017. Okay. So let me, let me, let me give you a little bit of counterpoint and then I want to get a reaction from Casey and Tom. Um, I think what, what some people might say, I haven't read any of the pro D site people and I don't spend a lot of time on it, but, um, I would assume, I would assume what some people would say was like, look, um,

You are speaking very much from within the current enterprise of how science is done. When you really break the rules, you don't really know what's possible. It's possible that, you know, some of this drug discovery stuff is happening outside the U S and maybe it's happening outside of the, you know, multiple fate, like the outside the FDA system for drug approval and drug discovery. Um, it's also possible that these things could get sold much earlier in the timeline relative to actually, you know, raising single handedly, raising all the money to take the thing to the finish line. Um, and, um,

You know, one of the things that we learned from DeFi is that there were a lot of bad ideas in the beginning and people wasted money and did stupid shit. But eventually, like collectively, they were able to learn and create better and better, more useful products over time. Like, why don't you allow the same thing to happen in DeFi that happened in these other domains? So, yeah, I guess I'll go through your criticisms one by one.

So the first the first concern of like, hey, people, oh, well, that's just the US. What about not the US? So a lot of drug companies actually do their drug trials in other countries that are cheaper or have weaker regulation first and generally get drugs passed to, you know, stage three in Madagascar much earlier than stage three in Western countries.

And they do that. That's how they kind of get like that sort of this like regulatory arbitrage that they do for like getting to first time in man quickly. And so it's already that's already being done. So I don't think that there's that much more efficiency you can get from that. And I'm not sure the decentralized part helps you that much.

The second thing, which is the risk transference aspect. This is where I think the D and D-Side is actually useful. This is why I think the prediction market is probably more valuable than the raise money for my drug. 50 years ago, an average small biotech company tried to exit by actually going public and having a drug that worked and their stock went up.

kind of recent years, most companies

compounds designed by smaller companies and smaller groups of people all exit to large companies because the marketing and distribution costs are too expensive. Running the trials is too expensive. And so like if you even look at like the vaccines, like why is it called the Pfizer vaccine? Pfizer didn't even fucking invent it. It's because it was too expensive to actually even get to production and get to get through all the regulatory costs of Pfizer, like did a joint venture. But like realistically, they own a lot large percentage of it.

And so there's kind of this thing where yes, there is actually some value. And like my old work, realistically, we spent like a billion dollars on building hardware, doing tests, whatever, over 15 years. And we sold like four compounds for like $500 million. So like, I mean, yes, maybe we get royalties or my old work gets royalties on them, whatever. But it's like people are doing those things. They are somewhat inefficient. And I think if you have better markets for like

the existing system, it's almost a better use of capital than speculating on... It's almost like saying, hey, I hope that you can train a foundation model on $1,000. If I told anyone that, they would laugh me out of the room. And that's kind of what the Adesai people sound like. On the other hand, like I said, there's things that could be good. But unfortunately, the

Did you have a third? You had the third point, right? But the last one was like, why don't you let people face plant and learn? That's how, that's what DeFi did. And DeFi turned out. So DeFi, this gets back to our thing about good metrics in DeFi. There was, there are actually good non-gameable or at least costly to gain metrics. And,

And whoever won that would take advantage of taking some market share away from the centralized equivalents of them. And that was a very clear value prop. Like your product works if you're able to design enough features that people want to move from centralized to decentralized. That's not clear here. Like there's no obvious, this has it even worse than the layer one problem of what do I spend my points on? Casey, as the frontier tech investor, what's your take on DeSci? Do you agree with Tarun or do you think he should lay off?

My take of what DeSci is today is pretty simple. It's that crypto people make a lot of money and they look for places to rotate it. And they rotate that into a variety of things. DeFi, meme coins. DeSci is just another vertical that they said, great, I can stay on chain and I can put money here and it might rip upwards. And I like the EV profile.

And I think it's that simple right now. I think the vast majority of DeSite, as Tarun said, is really just...

They're trying to get beta to science and they're trying to get access to it in a way that's on-chain. It actually really rhymes with what we're seeing in the AI space, where a lot of these tokens, people can't even differentiate between them. They're just saying, "Okay, this is AI on-chain and I want some exposure in my crypto portfolio to AI." Because AI is generally hard to get exposure to when we're in a video stock is.

I think it's mildly interesting, but I agree that the mechanisms are pretty basic and raw and it's not very inspiring to say the least. I do agree like the prediction market and I think that we'll see maybe even new mechanisms that come about that are more crypto native. And then I think the, I'm just like super bearish on DAOs. I've always been bearish on DAOs. I don't think they work. I think they're like a,

kind of outcome of there was like other things in crypto that have led to them. And I have a really hard time believing that that's the right format or structure for deploying money. And so I just don't think that's really a core feature. So any like the DAOs that are really singing the praises of that, I'm kind of against. Tom, what's your take?

Yeah, I mean, I think everything that Trude said could probably also be said about ICOs. And I think the important thing is like, it's not, it's sort of like an existence proof. It's not that this is the best way to,

fund startups or to fund projects. But it's like, yeah, you need, you know, one or two hits to sort of demonstrate that, yeah, maybe you can actually have this thing work. And like, I would say like Ethereum being a good example of that. But yeah, I mean, you have like, you know, no accountability, no real guarantees that like this token is going to do something. I don't think anyone is thinking that, yeah, we're going to raise

billions of dollars and, you know, take one of these drugs to market. But it's like, yeah, I mean, you also read about families that, you know, they're afflicted by some

you know, very niche illness and maybe there's some, you know, drug in the market or went off market and they want to fund it somehow. And this feels like, actually maybe not the most effective way to do it, but it's like, you know, why not kind of let the market feel it out versus, I also feel like kind of similar to ICOs. It's like the loser, the better. I think the token is IP stuff is also very stupid, but I think the idea of it's like, yeah, you, you know, buy this token or you put some money into a contract and like,

Some of it's going to get funded to something and hopefully it goes well and maybe the token will do well, kind of like DJT, Truth Social. It's like, are people buying that because they really want to be bullish Truth Social? No, they just want to support Trump and this is the way that they're expressing that. And I don't think you have to overanalyze the market structure around DeSci, although I think all the points maybe around the quality of participants, I'm going to defer to Tarun's judgment on.

I think one thing, agree with both of what you were saying, and I think Tarun brought up some really incredible points. I

I think that what we're going to see, just given the cohort of people that are in crypto today, I would be very surprised if DeSai is bringing in a new cohort. I think it's that the current cohort of people that own crypto are generally interested in these things because the ICP of it is interested in frontier tech because they're tech forward. And so one thing that I could really see materializing is insert new frontier tech vertical X crypto and that being another category. I actually think the neuro X crypto category might come to fruition in 2025. We'll see.

And so, yeah, I think this isn't limited to DeSci. And I think that it's just, I want to call out that I don't think it's a different cohort for what it's worth. Yeah, I think actually, Casey, your mental model of this really clicks for me. Like, if you look at the drugs that are on Pumped Out Science, they're all about longevity. Right?

Right. Which is actually a very narrow part of the drug market. Most drugs are about weight loss or, you know, baldness or, you know, some kind of sexual performance drug. Like that's those are by far the biggest markets for drugs. I think the largest market is still the cancer drugs, although GLP-1s are getting close. Is that right?

Yeah, cancer drugs have infinite. All right, got it, got it. I obviously know nothing about this. I'm judging from TV commercials. So yeah, and of course, longevity is just a tiny, tiny fraction of this, which of course intersects perfectly with what crypto people are interested in. So there is some sense in which this is not an entirely commercial enterprise. This is a, okay, you're super rich,

And you kind of want to play science and like put on your little lab coat and, you know, bring out your magnifying glass. And it's like, okay, cool. This is like your way to play science. And it's fun. I like that the meme coins are like cosplay. That's like basically what you're saying. It kind of is, right? And like it's kind of self-consciously so. Like literally there's science being launched on pump.fun. You know, like, okay, they're kind of, it is fun.

Not entirely tongue in cheek, but it's also not entirely self-serious, right? It's literally called pump.science. So there's a sense in which like people kind of know like, yeah, you buy in early and then you sell, you know, once it goes up 10X, you know? So like you're not really along for the ride. You're not really playing for like stage three. You're playing for like the memetic factor and it's inflected with this science-y kind of vibe. And it's a little bit like, you know, you go on Character AI and

And sometimes you wanna talk to Harry Potter and sometimes you wanna talk to Batman and sometimes you wanna, whatever. And it's a little bit like, okay, you wanna trade meme coins

Sometimes you want to trade science meme coins and sometimes you want to trade, you know, racist meme coins. And like, there's a different flavor for every day of the week or whatever you're feeling inside. That feels like a better explanation for what's happening right now in DeSci than that it is serving an underserved part of the market or that it's trying to displace funding mechanisms generally. I think it's plausible that someday it could get there. But the reality is that, you know, to Tarun's point,

biotech funding or drug discovery, pharma funding is very, very specialized.

Because it's extremely technical. It requires a lot of background knowledge. And it's exactly not the kind of thing that ordinary people off the street are investing into. And when they do, they're almost always tons of adverse selection. Look at the Cassava thing. So there's this Alzheimer's company that was down 90%. But it was like the meme stock favorite. And it was because no one kind of bothered reading the pretrial data. In fact, Martin Shkreli...

has a long diatribe about it like a week before the thing blew up and the stock was still going up so i i i agree with you that fine just call it science meme coins the d-side thing is a scam like the they're just science meme coins that's fine and whatever yeah okay so if they're just if they self-style as science beam coins then to ruin you're okay with it yeah whatever that's no different than whiff just just you know maybe get on the sphere instead this time

Got it. What if they're signed meme coins, but they do send some of the, if they ever get acquired, they send the revenue back to the token. Would you have a problem with that? Did we lose him? Tarun Raig quit. He didn't like ever saying kind words about Deesai. I think this thought experiment I was bringing up for him just made his brain break. And so he couldn't, he had to exit. Anyway, we're out of time. So we got to wrap. Thanks everybody. Thanks Casey for joining us. And we'll see you all next week. Ciao everybody.