Sam Atlman Reveals Remarkable Changes Coming To O1

So OpenAI had their dev day recently in London and at this event we actually got a few pieces of information from Sam Altman on the future of reasoning models. The reasoning models in question are of course, OpenAI's O1 series. The O1 series are far different to GPT4O, where these models can reason and plan long chains of events that work really well. And this is going to be something that we continually see see as we start to move towards this paradigm. So take a look because you actually genuinely want to pay attention to this because this is the new paradigm that we will be moving towards and there's a lot of focus on this area because this is how models will be in the future for any advanced kind of computation.

We want to make things better across the board, but this direction of reasoning models is of particular importance to us. I think reasoning will unlock, I hope.

Reasoning will unlock a lot of the.

Things we've been waiting years to do. And the ability for models like this to, for example, contribute to new science, help write a lot more very difficult code that I think can drive things forward to a significant degree. So you should expect rapid improvement in the OC system models and it's of great strategic importance to us.

So you can see here that he said this is something that is of course of great strategic importance to them, which means that at the top of their list is clearly going to be the O1 reasoning set of models. Like I said already, this is the new paradigm. This is where everyone's focus is going to go. GPT4O was great. It's an Omni model. ChatGPT4 was good, but this is the kind of reasoning that unlocks things that we previously weren't able to do. Like Sam Altman just said, they've been waiting years for this moment because it unlocks everything else that the future of AI has promised us so far. Things like in healthcare and research and science and agents. This is exactly what these reasoning models are going to be the base of. So it's really interesting also because Sam Altman gives us some insight to future model releases. He talks about, for example, 04, which would be the fourth iteration of the O1 model. And if you're an aspiring entrepreneur who's building any kind of AI startup, I urge you to watch the next few seconds of this clip because it tells you exactly where you should be putting your time and effort when it comes to an AI startup.

When you're a job right then that will not be as important in the future. If on the other Hand, you build a company that benefits from the model of getting better and better. If an Oracle told me today that O4 was going to be just absolutely incredible and do all of these things that right now feel impossible and you were happy about that, then maybe you were wrong. But at least that's what we're going for. And if instead you say, okay, there's this area where there are many, but you pick one of the many areas where O1 preview underperforms and 7 of patches just going to get it to work, then you're sort of assuming that the next turn of the model won't be as good as we think it will be. And that is the general philosophical message we try to get out to startups. Like we believe that we are on a pre a quite steep trajectory of improvement and that the current shortcomings of the models today will just be taken.

Care of by future generations.

And you know, I would encourage people to be alive.

Catch that. Basically what he said here is that look, these next kind of models are going to take care of many of the problems and issues that our models currently face. So if you're someone whose AI startup is trying to pick the low hanging fruit in AI, maybe you're trying to pick an area that OpenAI isn't focused on. This is something that is going to be ironed out in future releases. Sam Altman has previously talked about this because you get the OpenAI killed my startup meme. But of course just basically think about the O2 model or the O3 model as ironing out any kinds of issues. The kind of startup that you do want to build is one that is basically focused on providing another area or layer of distribution for OpenAI that they initially hadn't thought of. That's going to be the key thing for the future of building any kind of product or app or when these models are available for everyone.

I think fundamentally there are two strategies to build on AI right now or startups doing with AI. There's one strategy which is assume the model is not going to get better and then you kind of like build all these little things on top of it. And then there's another strategy which is build, assuming that Open Air is going to stay on the same rate of trajectory and the models are going to keep getting better at the same pace. It would seem to me that 95% of the world should be betting on the latter category, but a lot of the startups have been built in the former category. And then when we just do our fundamental job, which is make the model and its tooling better with every crank then you get. The OpenAI killed my startup meme, if you're building something on open on GPT4, that a reasonable observer would say if GPT5 is as much better as GPT4 over GPT3, was not because we don't like you, but just because we, like, have a mission, we're gonna steamroll you. But there's a giant set of startups where you benefit from GPT5 being way better. And if you build those and AI progress keeps going the way that we think it's going to go, I think on the most part, you'll be really, you'll be really, for the most part, you'll be really happy.

A developer or an entrepreneur who's building your own AI company. Just build in mind with the fact that GPT5,02,03, these models are going to be the ones that are largely better than they are now. And of course that means that if there are any limitations that you think, okay, OpenAI hasn't fixed this, I'm going to fix this and run off with this myself, I would say that that is a pretty bad bet. So make sure that your company is not doing that. And if you are wondering what the immediate updates to 01 are going to be, Sam Altman also released some really interesting information about 01. You can see right here that there was this image that we saw from Developer Day where we got to see five different features, one of them being a little bit more interesting than the others. But if we go down, I'll explain to you exactly what they are. So, of course, number one is function calling, which is where the model can connect with apps and websites to do things automatically, like looking up information or sending data to another program. And this is pretty useful because it saves time and makes it easier for apps to do things like show live weather updates or book a table at a restaurant without extra coding. Of course, there is then the developer messages. And this is basically where the model can talk directly with people who build apps and websites and giving them helpful information or feedback on how to make the app work better with AI. And this is useful because, you know, developers can fix or improve how their apps use AI faster and more easily. Now, of course, we're getting a new feature for A one called streaming. And basically, instead of waiting for the whole answer to be ready, the model can give you parts of the answer as it thinks of them. It's kind of like watching a video load in pieces instead of waiting for the whole thing to download first. And this is really useful because you get answers faster, especially when waiting for long or complicated responses. And then we have the structured outputs. Simple explanation being that this is just where the model can organize its answers in neat formats like lists, charts or tables instead of giving plain text. And this is useful because it makes the AI's output easier to use in other places. And like showing data in an app or exporting it to a spreadsheet without needing extra cleanup. And then of course, this is where things start to get really crazy because I didn't even know this was coming. But we get image understanding. So for those of you thinking that image understanding for 01 wasn't going to come, this is something that is going to probably come sooner than you think. And the way that Sam Altman actually spoke about this showed us that he's trying not to reveal all of the information. And it seems like something juicy is coming. With O1 image understanding, how will vision.

Capabilities scale with the inference time paradigm set by O1? Without spoiling anything, I would expect rapid progress in image. It's a bit of a camera plan.

Okay. And one of the craziest things that he actually said was that without spoiling anything, expect rapid progress in image based models. And one of the key pieces of research that I'm about to cover on my channel was something called simplifying, stabilizing and scaling continuous time consistency models. This is something that OpenAI have, you know, done research on and they've recently released a paper and they basically show how to, you know, go from diffusion models to this new method that literally you can create images in 0.1 seconds that's extremely high quality and it's at 10% of the compute cost for previous, you know, ways that we used to do it before. So I'm going to doing a full deep dive on this paper released by OpenAI just a few days ago. But this kind of has some crazy implications now for those of you who are thinking of about agents. I know agents are all the rage right now. Sam Altman also gives his view on what agents will be able to do.

The main example they seem to give fairly consistent is oh, you can like, you know, you can like ask the agent to go book you a restaurant reservation and either it can like use open pickle or it can like call a restaurant. Okay, sure, that's like a mild annoying thing to have to do and maybe like saves you some work. One of the things that I think is interesting is a world where.

You.

Can just do things that you wouldn't or couldn't do as a human. So what if instead of calling one restaurant to make a reservation, my agent would call me like 300 and figure out which one had the or some special thing available or whatever. And then you would say, well, that's like, really annoying if your agent is going to restart us. But if, if it's an agent answering each of those three places, then no problem. And it can be this like, massively parallel thing that a human can't do. So that's like a trivial example.

But.

There are these, like, limitations to human bandwidth that maybe Z agents won't have.