AI That Remembers: ChatGPT's New Upgrade
13:21 Share

OpenAI has just rolled out a brand new feature called Memory. And I think that this is going to be number one, a way for them to actually make more money. Number two, this is the moat that I think is going to keep them competitive and really keep everyone coming back to OpenAI when they're kind of starting to lose market share to other players. And finally, I want to break down exactly what Sam Alton, the CEO of OpenAI, has been

sharing about this. A lot of really interesting use cases that I personally can see. I'll show you a demo of exactly what that looks like if you're watching live. But otherwise,

let's get into what Sam Altman said. So in a recent post over on X, he said, we have greatly improved memory and chat GPT. It can now reference all of your past conversations. This is a surprisingly great feature IMO. And it points at something we're excited about AI systems that get to know you over your life and become extremely useful and personalized. Okay. I think the personalization element is probably like one of the most important. I want to get to that.

He then said, rolls out today for pro users and soon for plus users, except users in the EEA, UK, Switzerland, Norway, Iceland, and Lichestein. Shout out to everyone over in Lichestein that does not get access to this. I mean, it's

It's cool to get a shout out. He said, call me Lichtenstein. Let's work this out. He said, you can, of course, opt out of this or memory altogether. And you can use temporary chat if you have, if you want to have a conversation that won't use or affect memory. Okay. What is this? Because a lot of people are probably like, well, didn't they roll out like some sort of memory feature in the past? So in the past, ChatGPT had a feature that you could opt in and opt out of, which essentially said, if you said something really important, you could add like a note

to, you could add a note to chat GPT and be like, hey, remember this for the future. And it would have a little pop-up that would say like, hey, we're, you know, we grabbed this piece of information, we're remembering it. And in the, kind of in the back end of your account, in your settings, you could go look and see like all the things that it remembered. It kind of had this giant list of things that sort of remembered about you that were important. Usually things that you told it to remember about you. Now they're changing that. And so obviously that was something that I think would save a lot of money, right? They'd only have to remember a few things, kind of have this like,

little memory bank of stuff and probably as a way for them to test out this full rollout. The full rollout is where it remembers literally everything you have ever said to it in all of your chat history. A couple of reasons why I think this is important. Number one, we know that in like one big chat on ChatGPT, it's supposed to, it's sort of remembering everything, but you'll probably notice that if you get like 30 messages into something with ChatGPT, it starts to forget the top things. And that's because the context window

That essentially it's grabbing and sending to the AI model when it wants a response back.

It's like, look, this is like way too big for us to like send the entire last 30 questions and responses just to get context for like the question. Like you probably could get the context out of the last like 10 messages. So they sort of had this cutoff point and they definitely have the cutoff point with their API. So like all developers that are using this and adding it into things, but they are going away with that. And now they're just creating a, probably a rag model, but a giant repository of data on every conversation you have ever had with chat GP and everything you've ever said and all

All of that can be taken into consideration when you send a message. Now, typically this is going to be, you know, things that are like relevant will be queried against it. I'm assuming they have some like AI models to optimize and make this. So it's not like you're literally sending every single message.

chat message you've ever sent every single time. They're just trying to find the relevant pieces of it, but it's remembering and it's a drawing from everything. This is so fascinating. I want to show you a use case. Then we're going to dive into some more of the details going on here that I think are interesting. Okay. I just went to ChatGPT. And by the way, if you ever want to know like where this is, it's over in your settings on your ChatGPT account. There's a place called personalization and there is a toggle

called memory where it's you could just toggle on reference saved memories. They say let chat GPT save and use memories when responding. You could toggle this on and toggle it off. It's super easy. Okay, so I have it toggled on and I've had it toggled on. And by the way, this is going to come toggled on by default for everyone that has previously had chat GPT toggled on the kind of the like the debate their earlier version of memory. Now everyone that had that on is now going to get the full version. Okay, so

I said, what is a good business idea that we have talked about in the past? I'm like, I don't know what, you know, when I talked to Chai Chappie about in the past. Okay, the funny thing is,

I share my... I have like a work ChatGP account and I share it with all of the employees. I probably share it with my mom and my sister and like a bunch of people. If you want to use ChatGP for free, this is kind of like... This is my company account that like a million people are using and you just use it for free. I just want them to have premium pretty much. So it has all sorts of funny things on here that I don't remember ever asking about like women's health advocacy platforms. That probably wasn't a business idea I was asking about, but I'm assuming some other people...

One thing that was interesting is it's like you should do Christian music and wellness retreats. And it is true. My wife did just release a Christian music album a few days ago. So I've been asking questions about how to mix and master music and do different things and come up with different concepts.

And so, yeah, I've been asking it all sorts of questions about this. And now it's decided that I need to have a faith based workshop for self care and storytelling, whatever. I don't really honestly, the ideas, I don't think are fantastic. But what's interesting is it says, tapping into your wife's music and your shared values, you should do.

blah, blah, blah, blah, blah. It is true, right? So that is very interesting. This is just pulling it from some conversation I've had in the past. Now, if you'd done this like two days ago or yesterday even, you would not have gotten from like a brand new chat, you know, what are ideas we had in the past. You wouldn't have had it be able to pull up everything else. But now, it's actually able to pull up chats that I had in the past. So to me, yeah,

This is very, very interesting. And yeah, it's just a simple chat GPT prompt where I ask it questions and or ask it a question. And then it's pulling up conversations that I've had in the past. So this is this is really, really interesting. I wanted to TechCrunch just released a whole article about this. They said a couple of different things that were interesting. So apparently.

ChatGPT free users are not getting this right now. ChatGPT OpenAI said, quote, we are focused on the rollout to paid tiers for now. So if you don't have a premium account, and honestly, this is kind of how everything with ChatGPT and OpenAI goes whenever they release a new feature. Even when they're like two weeks later, the free tier gets it. If you want it on the cutting edge, you got to be a paid user for it, which it's 20 bucks a month. And I have it for a bunch of different features anyways.

What's interesting is they essentially are not copying Google. A lot of people might say that, but Google did just roll out a very similar feature to this in Google Gemini. And so this came out in February.

this year. So a couple months ago, and a lot of people, I think, don't use Google Gemini for a variety of reasons. I don't want to go and drag them too much. But recently, I tried out their new model, which is like above and beyond on all the benchmarks. It's like the best thing ever. But when I tried it, it like had some very obvious like hallucination fumbles in the first couple questions. And I was like, ah, it's just not going to become my daily driver yet. And I think Google has a lot of money, resources and talent. And it

will be amazing soon. Maybe it's there today. I don't know. It wasn't there two weeks ago when I tried it. But in any case, they rolled out this feature and it's kind of cool to see Google being the first one to roll out a big feature. And then immediately after we're having open AI kind of come and copy it a couple months later. So it feels very much like meta whenever, you know, Snapchat made the stories and then instantly meta goes and copies the stories and clones them and throws them into Instagram. It's like,

these models are all kind of competing in the same space right now. And so they're just kind of, if someone comes up with a good idea, everyone's going to try to jump on it. You have, you know, deep seek that came out with really showing you what the reasoning model was and what the thoughts were in chat. You kind of had a reasoning model, but they weren't showing you the thought process on it. And they immediately copied, cloned it. Now you kind of,

read the thought process, or especially like Grok does that now too, where they show you the thought process in detail. So these models are going back and forth with things. The reason why I think this one in particular is interesting is because this is the moat. This is literally the perfect thing that Chatubiki can do. We have a whole bunch of new AI companies coming to the forefront that are stealing market share. I'm going to be 100% honest. If you've ever used

grok.com grok is one that they just released a bunch of new usage metrics on the state of AI and grok has swooped up a whole bunch of market share it's xai Elon's AI company and they kind of branded themselves as like we'll give you answers to everything and to be honest like I actually it was my like daily driver for chat I'm using chat should be a little bit more now that they came with the image model and I'm like oh man this is super cool but

For most text responses, I've been using Grok because it doesn't really seem to... I don't know how to say it nicely, but it's like, it doesn't tell me like, hey, like, you can't ask that question or like, we can't respond to that or like, it doesn't really baby you as much. Like, if you have a question, I'm just like, hey, like...

I need to figure out this thing for my taxes and my accountant asked this question. Like, what's the, what's the best way that I can do X, Y, and Z? Like, you know, help me find like ways to save money on taxes or whatever. Like obviously legal stuff guys. But if I ask those kinds of questions to chat, it's like, Hey, I don't know. I'm like, whatever. It's like,

really strict and it's not a financial advisor, whatever, Grok will just kind of like give you all sorts of ideas. So that is why I like Grok. That is an illustrative example that I've used a handful of times, but it does that to kind of everything. So anyways, I've been using it more and the responses are pretty much the same quality.

And so, and actually the metrics showed that Grok was stealing a ton of market share. Now with this new feature that ChatGP is using with the memory, I think that they will be able to take back a ton of new users and the, or 10 of users that may have migrated other places. And the reason for that being is,

Everyone was using ChatGPT for the last like two years before these other models came out. Like a lot of people like Claude, a lot of people like, you know, Grok, maybe even Gemini is going to kind of come to the forefront now that they're beating them on a bunch of benchmarks. And so all the people that were really hardcore on ChatGPT because it was the first and the longest migrated away in the last, let's say, six months.

And if they go back, they have like a year and a half worth of data that's there. So I think that this is actually a good thing that they can use to pull back. And in no way is ChatGPT diminishing. We saw that after the most recent reports, after the image thing came out, it said there was 500 million users

weekly active users, I believe, or 400 million weekly active users. Absolutely insane number. So ChaiGP is still growing like mad, like crazy. But I think this is going to be a way for them to double down and stop the leakage out to other AI companies. So overall, this is very, very impressive. Some people have

you know, cited that they don't like this because they're worried it's going to be an infringement on their privacy. They don't want, you know, tech crunches saying that, you know, quote, not every user will be thrilled with the notion of opening. I vacuum up more of their information. Of course, fortunately, there's an opt out, whatever. Okay.

Yeah, if you're concerned about your data, I mean, you technically are giving it to them either way. It's on their servers, but they're not going to be used. They could use your data either way, I think, especially if you're a free tier. And the kind of the agreement is like, if you're a free tier and you're asking it questions, they can take that and use it to help train their models. So that's already how they're operating. And just because you now know and you can access your previous stuff, that actually doesn't really change anything. So I don't really think that's a good argument. Of course, if you don't like

the AI models, having your data, you can get a paid plan or an enterprise plan and they won't be training off of your inputs if you do that, that they've said. So that's kind of the way. And of course you're going to opt out of this, but I don't think opting in or opting out really changes much about how they use your data because you give it to them either way. But yeah,

In any case, major update from ChatGPT. I'll definitely keep you up to date on everything going down. I think this is huge, right? You can imagine if you say something like, I am allergic, like Allie Miller said on LinkedIn. If you say you're allergic to lavender and then you ask it for...

recommendations for a flower bouquet arrangement. It's not going to tell you lavender. It's going to say get daffodils or something else because it doesn't want you to die. And so I think that these are very good use cases. There's tons of good use cases. It can remind you of things that you may have forgotten about that you were talking about it to months ago. Maybe you're like, hey, I was working on this project and I can't remember what the answer was. What's the best thing? And it can go back and find the chats. Remember what you talked about with the project and what

the verdict was and let you know. And you can also build on everything you've done in the past. So I think this is phenomenal. I think this is memories and amazing moat and chat GPT and open AI have an absolute stranglehold on this because everyone was forced to use forced to, I mean, it was the coolest newest tool that everyone was using for the first year and a half seriously of the AI revolution. And so because they have all of that data, they can bring a lot of people back that want to keep access to all

all of that. So absolutely amazing. Really cool stuff going on. I'll definitely keep you up to date on everything happening in OpenAI and happening on ChatGPT and AI. Thanks so much for tuning in. I hope you have a fantastic rest of your day and I will catch you next time.