Open AI'S SECRET Project Sunshine Was Just REVEALED (GPT-4 With Memory + Advanced Features)

So, welcome back to the AI grid. And in this video, we're going to be talking about a secret project from OpenAI that is actually called Project Sunshine. But the more you watch this video, the more you're going to realize that Project Sunshine doesn't really have anything to do with Sunshine and is actually a new, upgraded version of ChatGPT that we might be getting in the very near future. Now, I know that many of us are very, very skeptical when it does come to leaks, but because, of course, leaks are just that. They're leaks, they're speculations, they're things that aren't 100% confirmed. Unless OpenAI comes out with a blog post with a tweet, or one of the creators of OpenAI actually says, look, this is what we're doing. But with that being said, there is definitely quite a lot of substantial information that you're going to want to see before dismissing this as pure hearsay. So, that being said, let's take a look at what exactly OpenAI's Project Sunshine is, who managed to dig up this information and find it, and exactly all of the real evidence on OpenAI's website that actually shows us that this has more credence than you might think. So you can see here that literally around two to three months ago, and this is something that I unfortunately did miss, someone called Tibor Blaho tweeted, interesting find in the ChatGPT public client side code hinting at Project Sunshine Beta features Sunshine. Try Project Sunshine, a version of ChatGPT with special capabilities. So what do you think its special capabilities might be? Share your ideas. So, around this time, nobody really had any sort of inkling what this version of ChatGPT might be. However, it's now come to light that this version of ChatGPT is going to be a new improvement on the existing version, with some really cool versions that aren't that crazy, but are definitely something that we do expect. So if we take a look here, you can see that ChatGPT's Project Sunshine is actually ChatGPT with memory, and it's a new GPT store layer. This guy says that ChatGPT's Project Sunshine, which was initially discovered in August 2023 as a version of ChatGPT with special capabilities, is actually ChatGPT with memory. The current version of ChatGPT features a new personalization section in its settings. Now, you might be thinking, okay, this is just one person on Twitter, this is the equivalent to one person like Jimmy Apples or Futurist flower, those previous OpenAI leakers. But this is a little different because you can actually check this out for yourself. So if you actually do check out OpenAI's JavaSCRI script on the same web page where the GPTs are, you actually do see this information. You actually do see this information immediately. Now there's actually several different tidbits that you can read here. And like I said, anyone can check the JavaScript themselves. I'm actually going to show you how you can actually check this when you go onto the GPTs page, because this is actually something that is currently on OpenAI's page and this is where the information is grabbed from this. Tons of Reddit posts about this where people have themselves actually check that this is a real thing. So even if this is some kind of secret file that OpenAI is hidden, it's still on their website and isn't just a screenshot from a random, you know, email or memo that supposedly leaked. This is definitely something that is on their server. So we can see here that it says memory onboarding modal.title or memory onboarding modal title. And it says your GPT can now learn from your chats. It also says keep the conversation going and your GPT will carry what it learns between the chats, allowing it to provide more relevant responses. And additionally it says improves over time and it also does say as your GPT will become more helpful remembering details and preferences. So let's actually go over some of these and what exactly they mean. So the first one, your GPT and now learn from your chat, actually means that of course, like I said before, your GPTs and I'm guessing your custom GPTs are going to be able to learn from your previous chats. Now, we aren't exactly sure how they're going to implement this, because we do know we right now, currently there are several issues with the custom GPTs, including prompt injection attacks that are currently working to perfection that allow users to extract data files, prompt information from that custom GPT. Which is why I'm pretty sure OpenAI hasn't pulled the trigger on the GPT store just yet, because boy oh boy, there are some real confidentiality issues with those things. Once those are ironed out, that's when we're going to get these features, I do hope, because as we know, privacy is a real concern. You can also see here it says that keep the conversation going. So I'm guessing these GPTs, after you have a conversation, what it might do is it might summarize what the conversation was about, add it to the memory bank of that GPT and then you can essentially have more relevant responses. And it says your GPT will carry what it learns between chats, allowing it to provide more relevant responses. Now this is something that has been done on some other versions in small scale scenarios, but nothing like this has truly been done before. Because as you, you do know every time you open a new chat that you know previously, what you talked about was forgotten. So this is going to be something that is really, really, really effective if it does work. And it also does say that it does improves over time, which is going to be good because as you enter more and more information to the GPT, it's going to improve to know exactly what you want, which means that custom GPTs in the future are going to be pretty crazy. And it does say as you chat, your GPT will become more helpful remembering details, preferences. So maybe where you work, maybe where you live, it's going to remember what you like, it's going to remember what you don't like. I think this is why also the GPT store is going to be delayed and rolling out, because this is definitely something they are trying to add. Because without this, this definitely isn't going to be as good, but this is going to set OpenAI once again apart all of the other large language model companies. This is something that they don't have. So if this is something that does come out, it's going to be really, really cool and really, really interesting to see exactly how it works. Then of course you can see there's another message that says manage what it remembers. And it does say your GPT has been designed to follow your instructions in chat and you can reset your GPT's memory or turn off these feature in settings. So it shows us that right here we do have a situation where we're going to be able to do this. Now, later on in the video I'm going to show you guys the guy actually accessing this feature in alpha or in beta. I'm not sure if some of you guys have this feature. I know it hasn't been rolled out worldwide, but I do know that it is something that is there. But of course, when I did just simply checked inspect element, you do see that the memory onboarding modal is still there and if you go through it, you can actually find this on OpenAI's page. Now essentially this guy on Twitter is basically breaking down this entire thing. He's basically saying your GPT will continually improve as you chat, picking up on details and preferences to modify what your GPT knows to teaching something new, you can literally ask it, what do you know about me? Where did we leave on our last project? Remember that I like concise responses, so it's definitely going to be something that is really useful. In addition, you can see here that of course the reset memory feature, so it says the current status of personalization is also displayed as memory off on the dashboard, indicating that ChatGPT won't remember anything you talked about in this conversation. Additionally, the new ChatGPT with memory feature is introduced via a new pop up and I'm going to show you guys that pop up in a moment. But the pop up basically says that your GPT can now learn from your chats, keep the conversation going. Your GPT will carry what it learns between chats, allowing it to provide more relevant responses Improves over time as your GPT will become more helpful remembering details and preferences and manage what it remembers. Your GPT has been designed to follow your instructions in chat and you can reset your GPT's memory or turn off this feature in settings. So like I said before, if you want improving over time, keeping the conversation going and manage what it remembers is definitely going to be something that most of us do find very useful and very effective. So with that being said, that definitely means that, you know, this is going to be really interesting to see how it is deployed because we do know that sometimes when features are deployed they aren't that good and they do improve over time. But with the custom GPTs, we do also know that many people are still waiting on the store and I'm going to show you guys a video of all this stuff in a minute. I'm just showing you guys a Twitter thread first. Of course we do have a really cool feature here too. It says the previously mentioned temporary chat is now visible in the model selector. However, it appears to be a bit buggy and alters the style of the prompt text input. Similar to the existing chat history and training on option in Settings, temporary chat also has its own announcement pop up explaining its functionality. So essentially temporary chat is basically where you can have a chat that doesn't remember anything at all. And of course I'm guessing that this is not going to be the main chat feature. The main chat feature is going to be one where we actually have our GPTs remember exactly what's going on. You can see here that he talks about a blank slate where your GPT won't be aware of previous conversations and it will still follow your custom instructions. If they are enabled, then of course we have no model training and of course temporary chats won't be used to improve our models. And it says that moreover, the GPT Store has received an update as it approaches its launch, featuring a prominent header, search bar and lists of my GPTs recently used, made by OpenAI. So essentially this is the video that this person released on Twitter. It shows you that this is the prompt, this is the pop up. And what's crazy about this pop up is that we did actually see this pop up before. So I actually did actually see this pop up floating around Reddit maybe a month or two ago. And actually, like I said, you can see it does say keep the conversation going, improves over time and of course manage what it remembers. Then of course you can see he goes over to chat GPT, he refreshes the page and then it says your GPT can now learn from your chat. So this is going to be something that is, you know, of course, really cool. Then of course you can see as well, when you do have this feature, this is the temporary chat feature. So like I said, it seems what we are getting from OpenAI is going to be GPTs that constantly remember what we're talking about and constantly improve based on the information that we give it. Which means that over time this could potentially lead us to the personalized AI scenario where we have AIs that constantly are learning about us, improve over time and become more and more personalized. And I know from my personal use that if my AI is able to constantly remember everything about my business, everything about my day to day, it's going to become a lot more useful because currently I have like 10 different chats open for certain things where you have to keep reminding it exactly what to do, the exact kind of prompts, where you're from, what you like, the tone, the style, the writing, everything. And of course GPTs was a kind of solution for that. I think what they're trying to do now on the back end is just make sure that this is more efficient, solve out the issues where there is the data leakage and privacy leakage, and of course make sure that you know, the things that they promise, for example, such as being able to sell these GPTs and monetize them is of course going to be quite effective because currently you have to remember that OpenAI isn't actually a profitable company despite being for profit. The $20 a month feature doesn't actually make them that much money. And you have to remember that their training runs are super, super expensive. I don't remember exactly how much it cost to train GPT4, but it was super, super expensive. Here you can see the temporary chat feature where it shows not in history. Temporary chats won't appear in your history. Your GPT won't remember anything you talk about and it does say that we may still keep a copy for up to 30 days. And it says blank state. Your GPT won't be aware of previous conversations, will still follow your custom instructions and no model training. I mean some people tick this, some people don't. But essentially, essentially just means that this is like an incognito tab for ChatGPT where you don't need to remember anything. You want to have a quick chat about something irrelevant. And like I said, it seems like we're moving more towards these personal AI systems. Then of course you can see that there is this button here where if you go on personalization you can see reset memory. So it says your GPT continually improves as you chat, picking up on details and preferences to tailor your responses to you, to modify what your GPT knows or to teach something new, just tell it what do you know about me? Where did we leave? Remember that I like concise responses. But of course you can also see that there is the button there to reset the memory. So maybe you entered some false information, maybe your GPT just knows where you live or maybe you don't want it to know all these things about you and you just want to start again. You could just click Reset Remember and it says your GPT will forget what it's learned from your previous chats. This can't be undone. And then of course you can also see that you can turn this on or off. So like I said, you can see memory off, memory on. It will be interesting to see how this does work, if this is quite effective and if this feature is going to be working or not, how this is going to be rolled out. Because some people as far as I know have something in alpha. Some people don't have access to this feature, but it definitely is something that is on the web page. I did check this out and I'm going to show you me checking out in a moment. And of course you can see right here that this is something from Futurist of Flower who has previously predicted this stuff with great accuracy before talking about how temporary chats won't appear in your history and your GPT won't remember anything you talk about yada yada yada. You, you know, that's basically the code again, of course this is something that we've seen before, your temporary chat history and then something that we did miss as well, something that I can't believe that we actually missed this feature. That's why I want to bring this up was the fact that Chat GPT was already adapting its knowledge base the more we talk about it. So this Reddit user around a month or two ago actually posted something about this. So they actually said, okay, that your primary GPT was will continually improve as you chat, picking up on details and preferences to tailor its response to you. Try saying things like, I moved to San Francisco in two weeks, always code in Python or forget everything about my last project. And you can see that it says ChatGPT adapts its knowledge base the more we talk about it. Now, the thing about this, this didn't blow up that much because I'm guessing the problem was, was that this was maybe rolled out to an only small amount or small subset of users, which is why most people didn't see this, including myself, including most people on Twitter. And the problem is that most people do see stuff on Twitter. It gets a lot of shares. It just honestly like gets shared on all the main AI Twitter pages. But this one slipped under the radar. So what I'm guessing is when GPTs was announced, they rolled out this feature and I guess you could say a beta or an alpha to, you know, a select few. And I'm guessing that this didn't make the final cut. I don't know what reasons it was, maybe they wanted to just test it to get some user feedback to see exactly how it worked. But you can see right here that this was something that was already there. Now also, so this is a source pace of ChatGPT and essentially when you click Control F and you click memory, you can see that there's memory onboarding right here. You can see if I just scroll down and I just keep scrolling, I keep looking, you can see it says memory onboarding model. I'm not sure if I can zoom in, but it's pretty hard to see. That's why I did take a thing of it. I'm going to scroll down again. Memory. Memory, yeah. Memory onboarding function. Memory onboarding modal. And you can see default message. Your GPT can now learn from your chats. Memory onboarding modal. Keep the conversation going. Memory onboarding modal. It improves over time and you can see memory onboarding modal manage what it remembers. So this is stuff that you can see that this is like literally I just opened up Chat GPT and went into view Source code. And this is what I'm seeing on the actual website. So maybe this feature doesn't get released, maybe they just scrap this feature completely, maybe they can't solve the memory problem. But right now this is actually on the web page. You can just click, you know, right click on any page, then click View Page Source. When you click that, you click another link like on the source page, one of them, and then this is exactly what pops up. So like I said, I'd be surprised if this feature doesn't come to pass because this is something that many people have wanted for quite some time and this is definitely something that would take GPT to the next level. And it's also something that we haven't seen other companies do. For example, Bard, we haven't seen any kind of these features, LLMs like anthropic open source AIs. We've seen nothing really like this before. There was one that I can't remember exactly what it was, but we haven't really seen anything like this before. Now there was another kind of tidbit that was kind of interesting, but I think this was a bit of a stretch from Jimmy Apples that said they have been sitting on this for a while and. And it actually references a tweet from Adam GPT where he tweeted, you are my sunshine, my only sunshine, in response to the tweet where this was called Project Sunshine, a secret version of OpenAI's ChatGPT. So I think that this is a bit of a stretch because this is like a popular phrase that some people do and Adam GPT does tweet a lot of different Dall E3 images. But again, we're never sure with as to if they're teasing it, if they're just tweeting about stuff like this. But I do agree with the statement that only with proper competition will they actually push stuff out the door. But at the same time, I do always want to make sure we are focusing on AI safety because with yesterday's video we did see that ChatGPT GPT 4, like the security and the safety aspects of AI are really, really needed. So I really don't want there to be this culture of AI companies forcing out the software just because another company is going to release something that could potentially be better. I want to make sure hopefully that these companies do focus on safety because all it takes is one major event for AI systems to get set back quite a bit. Because if an AI system is ever used to, you know, perform something that manifests itself in real life and has real world impacts that affects millions of people. We are going to be seeing tons and tons of regulation with this amazing software. So I do hope the AI safety is taken seriously. Now question is to you, what do you think about this? Do you think that this is going to be good? Do you think this is going to be bad? Are you still excited for GPTs? Are you thinking why is it not out yet? Like I said, the GPTs is facing some delays as we do know, because they have talked about it for a while and remember, of course OpenAI did just implode the other day. So with that being said, I think this is a great feature that should be added. Let me know what you guys think about it. All links to the stuff are going to be of course, as always in the description and your support is always appreciated.