GPT-4's New "Memory" Feature Is Stunning (ChatGPT Memory)

There has been some several secret features that have been added to ChatGPT that many people haven't actually realized. In today's video we'll do a quick deep dive on these features and why you should be paying attention because this secret update is a lot bigger than most people do realize and when you hear Sam Altman's conversation, you're going to understand exactly why. So one of the tweets that I saw recently that was fascinating was this tweet by testing a catalog and essentially it tweets about the personalization aspect of ChatGPT. Now if you're familiar with this, you'll know that previously there was a slightly different user interface for this and it was a different Xbox. But there is also a new text box where you can see in the personalization tab. Now I must add that this personalization is only rolling out to a select few because when OpenAI tends to test this things, they test it out in certain batches and they do it randomly. So I don't think there's a specific way you can sign up. Sometimes they do have the ability for you to test beta features, but features like this that are quite rare and haven't fully been fleshed out yet do randomly appear in some people's accounts and of course do appear in sometimes in other accounts. So you can see here that it says improve responses with your chats. ChatGPT will become more helpful as you chat, picking up on details and preferences to tailor its responses to you. To understand what chatgpt remembers or to teach it something new, just say, remember that I like concise responses. I just got a puppy and what do you remember about me and where did we leave on my last project? Now if you remember recently there was a new update after the GPT Store launch and there was a very small blog post about this feature. Now I can't find the exact post, but essentially what they are doing is they are actually rolling out this feature because there were tweets by like Greg Brookman and some other open AI employees that do confirm that this is 100% a real thing. So it's not something that is going to be going away because some of the comments made by Sam Altman in a recent interview with Bill Gates prove that this is the gradual update that they were talking about and this is where they're going to be leading towards Now. Something additionally that I did see from this screenshot that I will leave a link to in the description below is of course the Manage Memory feature. So you can see right Here that it says, as you chat with ChatGPT, you can. The details and preferences it remembers will be shown here. So it seems like we're going to be able to see exactly the details that ChatGPT is managing to pick up on us. And that means that we're going to be able to see just how much information ChatGPT is remembering from our conversations and what kind of things it thinks about us. Because if it does get some things wrong, we could then go ahead and delete them and of course edit them. So I think being able to visualize what this AI system knows about you is very, very good because it allows you to see if it's correctly, if it manages to remember certain things based on certain conversations, or if it manages to even infer certain things from your conversations. And I think that would be an interesting concept as well. Now, if you're confused about this memory thing and you're thinking, wait a minute, I've seen this before, I've actually seen the memory thing before. This is because likely what you've seen is an earlier iteration of this screenshot. So I have this tweet from Teba Blaho, and this was a screech, a video recording of what we saw from an early iteration from OpenAI in terms of this memory management system. I'm not sure if they're using this at the same time now, but in November we did get to see something like this. And you can see that there are several things that the user is currently able to do from this system. Now, like I said, I think I did have this for a moment, but it has now disappeared. And we're now starting to see this different user interface rolled out. But of course you can see this personalization area has been changed a little bit. And I think this is why we're still getting delays on the next model, whether it be GPT 4.5 or GPT 5, because certain things like memory that we currently are seeing on screen now are currently being just waited out. And we're seeing what is of course, the most effective method of doing this, because there are many different things that we can do with the system. So I'm guessing they're waiting to see what is the more effective. And of course you can see right here, the temporary chat, not in history, blank state, and no model training is something that was there before. Now, this was kind of fascinating because even before this leak, there was essentially someone that just viewed the source code of the web page and then we were essentially able to see that. And I Think that that is kind of interesting that ChatGPT and OpenAI do kind of add these things in. But one thing that I do want to show you guys before I continue talking about all of these new memory updates is this interview. Because in this interview, and you may have seen this interview, but there is a specific clip in which Sam Altman talks about this, the next two years of ChatGPT. And the reason this video is really important is because this shows us where we're going for ChatGPT. And at the time I didn't connect the dots in this interview that he did with Bill Gates. But now it makes more sense because in this interview he basically says that, look, we're leading towards personalization where it's going to remember stuff and be much more customizable. And then of course, we do have things like personalization being rolled out. So it's clear that although the name might not be changed from GPT4 to GPT 4.5 just yet, it's clear that smaller and smaller iterations are being loaded onto this system as a gradual update. But take a listen to this clip because I think it's really important for us to listen to Sam. And then of course I will show you guys some more features.

You know, when you look at the next two years, what, what do you think some of the key milestones will be? Multimodality will definitely be important. We started speech in, speech out, speech in, speech out, images, eventually video. Clearly people really want that. We launched images and audio and it had a much stronger response than we expected. We'll be able to push that much further. But maybe the most important areas of progress will be around reasoning ability. Right now, GPT4 can reason in only extremely limited ways. And also reliability. You know, if you, if you ask GPT4 most questions, 10,000 times one of those 10,000 is probably pretty good, but it doesn't always know which one. And you'd like to get the best response of 10,000 each time. So that'll be that, that, that increase in reliability will be important. Customizability and personalization will also be very important. People want very different, very different things out of GPT4. Different styles, you know, different sets of assumptions. We'll make all that possible. And then also the ability to have it use your own data. So the ability to know about you, your email, your calendar, how you like appointments booked, connected to other outside data sources, all of that, those will be some of the most important.

So you can see here that that is pretty much exactly what we're seeing and I think now that we've seen that, you know, the personalization is being changed and it's being upgraded to, you know, having your memory being managed and of course you say these very simple commands and it's able to remember them. This shows us that OpenAI are moving very quickly. Now one thing that you might need to understand from the new releases from OpenAI is that I do think that these updates and these changes need to be paid attention to because what most people are doing with OpenAI and their services and in terms of looking at the updates, is that people will wait for an entirely new model to be released. But I think that what OpenAI are doing is very smart because they know that if they release GPT5 and it's a shocker in terms of capability, it's going to cause some, I guess you could say potentially some societal unrest. But what Sam Altman has talked about was the fact that they need to release things very, very gradually. And of course adding things like managing your memory and the personalization is a very good way to do that. So this is where Sam Altman talks about the fact that we're going to be basically gradual kind of updates.

One of the things that we believe very deeply is that society and this technology have to co evolve it it. We believe in iterative deployment a lot for the obvious reason, which is that people need time to gradually update and think and figure out what the rules should be. But there's another part which is it's it the. You can't separate the technology from the world. You can't just, even if you get everything magically right, you can't build it in secret and then put it in the world all at once because the world is going to keep changing with each iteration. Which means on those middle cases I don't know what the right answer is yet, nor does anyone, because no one has really liked thought through and seen how the, the institutions and the world and society shift and reshape in, in response to this. So there'll be a lot of things that we'll have to start slowly on and iterate as we go and there'll be a lot of middle cases. But we do want to support the U.S. government and other governments too. And like I find, yeah, you can.

See that this is why we know that the iterative deployment is going to be a thing. So one thing that I'm going to be doing now is I'm going to be making sure that I do take a look at every single thing that is going on with ChatGPT in terms of these small updates. So one way that I'm going to be doing that is I'm going to be making sure that I scour Twitter and Reddit to see if there are any instances of new things just slowly being rolled out. Because this stuff is fascinating in terms of it shows us the entire direction that we're moving. Because like I said before, it was very evident that the ChatGPT system was just a base test for a very larger AI system and ecosystem that we're going to be moving into. So I think using the GPT4 system as it is in the future is going to be a thing of the past. And in fact, we're going to have completely personalized versions. And that's what we've had a little bit with this GPT system. And I think this is also going to be quite similar to MEM GPT, which is essentially a personalized AI with infinite context windows. And essentially, MEM GPT was something that was devised to enhance the performance of large language models by introducing a more advanced management scheme in terms of the memory, hoping to overcome the challenges posed by fixed context windows. And essentially it was basically giving the AI system the ability to manage its own memory, having a virtual context management and having interruption handling. So I think something like this has inspired the next wave. And of course, like I said before, with how systems are evolving and the kind of reception that we did see Rabbit's R1 device get, I think we all know that personalized AI systems, if they are done correctly, will be even better than a simple chat bot that we can just, you know, talk to and then just get a simple response out from. So now if you do want to be able to test out these features when they do go live, what you want to be able to make sure is, number one, that you are subscribed to chat GPT's plus tier. So you want to make sure that you upgrade your planet to the $20 a month tier, because that is the only tier where you do get access to experimental features that may change. For example, if you then go into your settings and then you go to beta features, you'll see this tab, and essentially this tab says that as a plus user, enjoy early access to experimental features, experimental new features which may change during development. And I think that's what we're seeing. So I likely think that the roadmap for OpenAI is to simply test out all of this stuff and I guess to add a more balanced output in terms of the memory and in terms of the personalization to test these things before we do get a future release, whether it be GPT 4, 5 or whether it be GPT 5, I think that that memory aspect is where we're going to be heading towards overall in this niche. Now, for those of you who are familiar with MEM GPT, you'll understand and know how the system works. But for those of you who aren't familiar with this new style of update that we're going to be getting with ChatGPT, I think it's important for me to cover some of the demonstrations of how this system updates its memory and how it changes its interactions with the user based on this memory GPT paper. So in the paper we can see that MEM GPT says, hello, Chad, welcome. I'm excited to embark on this journey with you. As a PhD student in computer science, I can only imagine the fascinating ideas you're juggling. Can you share what you're currently working on? I'm all ears. And then you can see I took the day off today. My mom, Brenda, baked me a birthday cake. It was my favorite, chocolate lava. And then you can see it updates its context to say birthday, the 11th of October. Its favorite cake is chocolate lava made by its mama, Brenda. And of course you can see that it, you know, it just continues on the conversation trying to get out some small information. And I think this is likely where these AI systems are going to go in terms of being much more useful. Because the very big problem that we do have with these AI systems is that every time we have to get a conversation out, we have to tailor it specifically to our user interests. And that is going to make these AI systems so much more powerful. Because the only limitation that I've truly seen from these models in terms of getting the maximum output that you can from these current systems is the fact that people don't know how to prompt them properly and they don't know how to give it all of the correct information that gives it the context that will provide you with the type of information that will help you succeed and whatever it is, whatever kind of task you're doing. And then of course, you can see here it says, hello, Chad, it's a pleasure to finally have a conversation with you. I understand that you're studying computer science and have a keen interest in dynamic areas like Formula one and sailing, what draws you to those interests? And it says speed, of course, at thrill, the adrenaline. And then here he says, the conversation history will soon reach its maximum length and be trimmed. Make sure to save any important information from the conversation to your memory. And then of course it actually manages to do that. So it says key personality trait, enjoys high speed adrenaline rush activities, likes Formula one racing and intense gaming sessions in CS Go. And another thing you can see here is that this is how it actually updates. So for example, it says welcome back fancier chat about horror movies. And he says actually I don't like horror movies, I'm more into romantic comedy. And and then the system essentially says it's going to update its memory saying I watch horror movies. Replace this section with I liked romantic movies. So the system is able to learn about how you change over time and it says, you know, apologies for the mix up earlier, yada yada yada. So you can see here that MEM GPT is essentially, you know, conversational. Like agents are virtual companions and personalized assistants aim to engage users in natural long term interactions potentially spanning weeks, months or even years. And I think this kind of paper really goes to show us what we're going to be getting from these future models. And I think think everyone is overlooking this personalization feature because it's not some big announcement. But I think ChatGPT and OpenAI know that whilst the GPT Store is very good, personalization is just something that you can't get away from because the personalized nature of our friends is what makes the connection so good. Because we have all these unique shared experiences and memories in a personalized AI system that's able to get deeply ingrained in terms of remembering key details about your life. Like your birthday is going to be something that is far more valuable to you.