OpenAI's SECRET AI Agents Are ALREADY HERE!

So there was actually a recent article by a pretty trusted source within the AI industry that actually gives some key details about OpenAI's secret internal AI agents and internal AI research assistants. This stuff is genuinely fascinating. So make sure you fasten your seatbelts because this is going to be a wild ride. The information have been around for quite some time and they've been giving us really nice pieces of information when it comes to the secrets going on at OpenAI. And recently they said that OpenAI in duel with anthropic doubles down on AI that writes software. So essentially this article is basically talking about the really intriguing things that goes on behind the scenes at OpenAI's fight for competition. And trust me, once you watch this entire video it going to see that this thing is absolutely insane. So we do have the fact that ChatGPT, the OpenAI product, is actually facing fierce competition from rival artificial intelligence startup anthropic. And OpenAI is putting more effort into improving the tools that it offers for software programming. Some products or features under development aim to make it easier with to use OpenAI's AI for coding tasks inside major code editing programs like Microsoft Visual Studio Code, while others aim to take on bigger software development tasks. Now this isn't the main piece of news. All this does is set the stage for this. It says here, for instance, OpenAI has worked on a product to handle software engineering tasks that might take a human hours or days and to automatically write and execute code for complex applications based on customers instructions. Said a person who has heard from OpenAI leaders about it. So it's clear from this that OpenAI has an internal product that can handle software engineering tasks that might take human hours or days. Okay, that is pretty crazy. They're basically saying that this is the kind of tool that can save the average human hours or days. And this article gets even in more detail. We can see here that it says and it automatically writes and executes code for complex applications based on customers instructions. So this seems to be some tightly kept secret within the walls of OpenAI. But we're starting to see the signs that this is really advanced. Now OpenAI probably has been working on this for quite some time, but by the looks of things, a product that can handle software engineering tasks that might take a human hours or days does seem to be pretty advanced compared to what we traditionally have on the market. You might be thinking okay, so this is a product that might be coming soon, but when can we expect this kind of product? You can see here it says when and it isn't clear when OpenAI would launch such a product. Now there's a lot more information in this article it talks about the fact that now OpenAI is somewhat on its back foot in coding because the rival AI startup Anthropic has become a formidable competitor. And on Tuesday, Anthropic announced a new software that can use computers the way humans do to take actions on behalf of people, such as moving a cursor, clicking buttons and typing text. Now this is true because Anthropic did release this technology and it is really cool with us to what it can do. But it seems that this intensifying competition means that OpenAI is in quite a rough spot. Now OpenAI, it does say that they've also been developing a similar product known as a computer using Agent for months, but haven't managed to launch this. And this is something that in the other article it's described as pretty much exactly what Anthropics is. Now I'm not exactly sure the key details of that kind of agent because the article that it references is actually from quite a while ago. You can see that this article is from February 7, 2024. And you can see it talks about how OpenAI is developing a form of agent software to automate complex tasks by effectively taking over a customer's device. And of course, this is one of the things that OpenAI will have to temper fears about agents that access users computers, given that such a takeover may remind some people of malware, which is known to illicitly seize control of computers and steal their data. And it says that OpenAI plans for the agent that takes over people's computers to to require the user's permission to work and to operate in a personalized fashion and respond quickly the way that Apple does on the phone. And the craziest thing about this is that OpenAI did talk about how the fact that this could have been the project that will change everything. So it's pretty crazy what OpenAI are already working on. So I do think if OpenAI have been working on this product for months, but haven't managed to launch this yet, I just think that this product simply isn't ready for a worldwide public release because there hasn't been any real announcement. It's just been teasers and leaks from various forums and people like Jimmy Apples. Now one of the craziest things about this, and there's still a lot more information to get through, is the fact that Anthropic are actually ahead according to this article. Okay, it says by some metrics that OpenAI uses to evaluate models coding abilities. Anthropic models recently had an edge, according to a person who Talked to senior OpenAI leaders about the issue. So this basically means that it might seem that apparently OpenAI are behind in terms of the coding aspect for what these models are capable of doing. And if you've been paying attention to the ecosystem, you'll know that the majority of people aren't using GPT 404 code, they're using Claude 3.5 sonnets of code. And that has had people achieving remarkable results with their projects and their products, and has led to an entire ecosystem of people that are now building low code or no code small SaaS products to many different audiences. And I think this statement is wild because for the first time we're starting to see that OpenAI isn't in front of one of the main areas of development. Now, I don't think this is as bad because when you think about it, coding is a small subset of the various different abilities. And OpenAI did recently have the O1 model. So I wouldn't be surprised if this isn't something that doesn't concern OpenAI too much. But Anthropic models recently having an edge is genuinely quite surprising. I would have suspected the GPT5 or whatever model is coming next from the GPT series would have likely been a lot better, but I've used personally the 01 model for coding and it just doesn't seem that great. Now, the crazy thing about Claude is that apparently CLAUDE is so good that it's boosted the developer's productivity to the point where Anthropic might actually alter its hiring plans. If you haven't heard recently, Claude, Anthropic are going to be hiring some people. But they were like, you know what, let's actually see how things play out. Because this new tool that we've released has actually boosted the productivity of these developers so much that maybe we might not need any new hires at all. And Anthropic has projected that by the end of 2024 it would generate revenue at a pace of $1 billion annually, or $83 million per month. And that's 25% higher than the initial projection at the start of the year, which is absolutely crazy. Now, of course, the co founder Daniela Amodei, told the information that AI coding can't fully replace engineers and still need some coding, but it has boosted the productivity. Now, the craziest thing about this is that some of these leaks, there's actually some information to support them and this is where we get into the meat of the video. This is where we get into the internal research assistant. So it says OpenAI's internal tools show the broad potential of its AI for coding related tasks. And the company, this is where it gets interesting, has developed an internal research assistant tool to help AI researchers work faster, including by generating code for their experiments related to AI models, according to an employee. And the craziest thing about this product is that this product is apparently popular internally, this person said. But it's unclear whether the company would ever release it publicly. So there we have it, guys. There is an internal tool that apparently OpenAI are using to to accelerate their workflows. And it's an internal research assistant that helps researchers work even faster by generating code. So when we think about the fact that there is an internal AI assistant, AI agent that is able to help AI researchers and it's a popular product internally, this means that we're quite further along than you do think. Having AI that can conduct AI research, even if it's assisting researchers, is a pretty incredible feat. Now, I'm not sure if this is related to the O1 model. Perhaps it could be a specialized version. But there are some key details on the Internet that do show that this isn't just a crazy leak. Now, I remember, I think it was six or seven months ago. I was making a video and Jimmy Apples tweeted something that was quite vague. He tweeted this, the new leaks of the internal OpenAI subdomains. Now, at the time, I didn't really know exactly what this was, but you can see in the highlighted areas, we can see that there is a bunch of internal assistants on these OpenAI subdomains. You can see that you've got the assistant Scientist Internal At OpenAI, we've got the health staging scientist, and we've got the engine scientist. So it seems like there are many different kinds of models that OpenAI has had since 2024. Slash, mid to late 2024. That gives us the idea that they are using some kind of model to aid researchers in some kind of work. I think this thing is absolutely incredible. Slash absolutely insane. But I think that now, looking back on it, Jimmy Apples was clearly onto something. Maybe he couldn't tweet out exactly what was going on, but now that we have the information, from the information, we can see that this is an internal research assistant that helps AI researchers work faster. And what happens when that internal AI research assistant is able to be fully automated for a period of time? Things are about to speed up quite a lot. Now you can see here that it talks about this internal tool and it says that it appears to be a step towards the development of AI that can conduct research on its own, a skill that requires far more than coding, including generating ideas for new experiments. And OpenAI leaders have publicly said that this was likely to come in the coming years, which they have. So apparently. Well, I wouldn't say apparently, but this is of course one of the major steps towards AI that can conduct AI research on its own. We've all seen the Sakana AI research scientists, but this seems to be along the lines of something that is a lot more promising. Now of course we did also have information about a product that OpenAI were considering developing. They have also considered developing a version of these companies Canvas product, a new ChatGPT interface that makes it easier for users to draft documents and code. And they're thinking about making a specialized code version of this that can integrate with developers code repositories, according to a person who has spoken to OpenAI managers about the idea. So if you aren't familiar with Canvas, it's basically ChatGPT, but you can actually use it to write long form documents and when writing and editing code in that thing, it is actually a really pleasant experience. So OpenAI might develop a singular product that could be just code based on that. Now the craziest thing about this is that they've actually managed to internally demonstrate a preliminary version of an agent that would use the computer to do tasks like order food for delivery, according to a person who has seen the demonstration. So for those of you thinking that OpenAI don't have agents, it's clear that they already have them and it can already do things like order food and.

I'm guessing it can do other tasks. Now there was also another recent demo which I think you all should see and it wouldn't make sense for these kinds of agents to be demoed now since we actually have the kind of model that allows long form reasoning, which is of course the O1 model. So recently at an OpenAI dev day they did a short demo which they showcased exactly how these models work when they do work as agents.

Could you place a call and see.

If you could get us 400 strawberries delivered to the venue? But please keep that under fifteen hundred dollars.

I'm on it.

We'll get those strawberries delivered for you.

Hello.

Hi there, Is this illness I'm Romance AI assistant calling about it. Fantastic. Could you tell me what flavors of strawberry dips you Have?

Yeah, we have chocolate vanilla and we have peanut butter.

Wait, how much would 400 chocolate covered strawberries got? 400?

Are you sure you want 400?

Yes, 400 chocolate covered strawberries, please. How much would that be?

I think that'll be around like $1,415.92.

Awesome. Let's go ahead. Place the orange. 400 chocolate covered strawberries.

Great.

Where would you like that delivered?

Please deliver them to the Gateway Pavilion at Fort Mason and I'll be paying in cash.

Okay, sweet.

So just to confirm, you want 400 chocolate covered strawberries to the Gateway Pavilion?

Yes, that's perfect. And when can we expect delivery?

Well, you guys are right nearby, so it'll be like, I don't know, 37 seconds.

That's incredibly fast.

Cool.

You too.

So with the rise of 01, we do know that agentic capabilities are literally just around the corner. And I think 2025 is going to be absolutely surprising. Now, if OpenAI are developing these kinds of products, ones that can code internal agents, I think they definitely have a lot up their sleeve that most people can't even fathom. And that's probably why it was only a few days ago when the CFO said every week her mind is constantly getting blown. Now, the crazy thing about this is that the article also speaks about overlapping products. So it says that the products that are coming next and the products that are under development are could threaten customers of its application programming interface business who sell coding assistants. And it says among them is Cursor, which actually uses OpenAI models for its product and has taken venture capital from OpenAI. So OpenAI's future coding products may overlap with open source coding assistants such as Continue or SWE Agent, which uses LLMs to help users write code and fix bugs. So overall, we can see that it's quite likely that OpenAI is developing their own internal autonomous software engineers to get them to do work for them. We can see that they're able to do work that might take a human hours or days and automatically write and execute code. We could also see that they've got this internal research agent that helps researchers work faster by generating code for their experiments related to AI models. And according to an employee, and of course, we've seen that there are internal subdomain log details of these websites that show us an internal scientist, a health staging scientist, and some kind of engine. So overall, it seems like we're going pretty fast if we think about what OpenAI has behind closed doors. And if you do remember, it was on the GPT6 trademark. We saw that OpenAI is going to focus on agents. So with that being said, let me know what you think about this stuff. I would love to know your thoughts about OpenAI's internal use of the models, because it doesn't seem like we're going to be getting some of these products.