Sam Altman Finally Reveals What ASI Will Be Like! (Artificial Super Intelligence)

A fascinating interview because for the first time he actually talks at length about super intelligence. Now, super intelligence is something that we don't frequently get talks about because essentially that's something that's often seen as something that is too far into the future. We can't even imagine what it looks like. But recently there have been more and more claims around the industry and whispers, so to speak, about potentially super intelligence coming within a few thousand days. Now this article and interview is just completely fascinating because not only do we get some Altman statements, but we also do get the statement from a former OpenAI researcher who actually gives us a little bit of information about superintelligence as well. So you can see right here, the article is titled Sam Altman, AI is Integrated Super Intelligence is Coming. And then of course he talks about the fact that he was interviewed and he talks about the general state of artificial intelligence in their world. Now, one of the things they actually asked Sam Altman about was the September Manifesto. And if you are not familiar with the September Manifesto, trust me, this is something that you should read because I would say that this is potentially the best look at how the future is going to be. With all the rapid advancements in AI, it's not often that we do get a blog post that is so detailed, yet so simple to read and understand, where we can truly understand and how society is going to be moving forward. So what they're referring to is this called the Intelligence Age. This was published not that long ago, but it was something that I found to be very insightful for looking forward into the future and just understanding how society and everyone is going to be shaped by AI, AGI and of course asi. Now, one of the things that he did say here was that imagining what things will look like in an 18 months, you know, as we round out the summer of 2026. And how does super intelligence emerge? And he says you have to look at the rate of scientific progress, he said, describing how things might compound advances over the next few years. And this is really interesting because when we do take a look at superintelligence, one of the key fields that superintelligence will impact the most is of course, is scientific progress. That is the area where superintelligence is likely to have the biggest impact. Of course, you've got mathematics, which is of course something that we do know has been rapidly advancing with the AI models we've seen with Test Time Compute, how much these models are now scoring on benchmarks, which is just simply incredible considering it's only the second iteration of the model. Of course, we've got science questions, which is going to be really fascinating for those who work in scientific fields, because when you have an AI that you know, you can essentially say, okay, I need you to go run this test or do that for me, that is going to be something that just has a remarkable level of unlock in terms of not only productivity, but also the scalability of research. I think one of the things that people are discounting about AI is that maybe the AI doesn't get to 10 times smarter than us. But even if it doesn't, imagine you could, let's say, clone 10,000 scientific companies and tell them all to do scientific research. Imagine how quickly things could get done. Of course, there are delays because of supply chains and physical things, but I still think that in terms of pure software development, those areas are going to move completely at light speed because there aren't any real bottlenecks to how quickly you can move. And he actually talks about this in this short snippet right here.

One thing that I use as a sort of my attempt at my own mental framework for it is the rate of scientific progress. If the rate of scientific progress that's happening in the world as a whole tripled, maybe even like 10x, the discoveries that we used to expect to take 10 years, and the technological progress that we used to expect to take 10 years, if that happened every year, and then we compounded on that, the next one and the next one and the next one, that to me would feel like superintelligence had arrived. And it would, I think, in many ways change the way that society, the economy work. What it won't change, and I think a lot of the sort of AI commentators get this wrong, is it won't change like the deep, fundamental human drives. And so in that sense, you know, we've been through many technological revolutions before. Things that we tend to care about and what, what drive all of us, I think change very little, or maybe not at all through most of those. But the world in which we exist will change a lot.

Now, what's crazy about all of this is that those of you who are wondering when super intelligence, as many people are still asking, when we're going to get AGI. One of the key things that he actually stated in his blog post, or the September Manifesto, as Forbes calls it, is essentially the fact that, you know, it is possible that we could have superintelligence in a few thousand days and that may take longer, but he's confident that we'll get There. So this is going to be something that is super fascinating because, of course, a few thousand days is a varied timeline. It does leave a lot up to speculation. But he actually does go into a little bit more detail in terms of the actual timeline. So for those of you who wanted a date, I'm not sure if you remember this interview that he did where he actually talks about it could be 3,500 days away, which is a little bit later than you might be initially expecting. But when we think about the level of impact that that kind of technology would have, 3,500 days for a technology that could. You know, when we think about how discoveries compound on another, like the invention of computer led to society being transformed, and then, because now society is more connected, we've got other things going on. I mean, it really is fascinating when we really try and think about how the future is going to be.

In the essay, you actually say a really big thing, which is asi Super Intelligence is actually thousands of days away.

Maybe.

I mean, that's our hope, I guess, whatever. But that's a very wild statement. Yeah, tell us about it. I mean, that's. That's big. That is really big. I can see a path where the work we are doing just keeps compounding. And the rate of progress we've made over the last three years continues for the next three or six or nine or whatever. You know, nine years would be like 3,500 days or whatever. If we can keep this rate of improvement or even increase it, like, that system will be quite capable of doing.

And for the individuals who would potentially call Sam Altman the speculator or someone that, you know, is drunk driving up hype to get more people to invest in his company, essentially, this is no longer something that Sam Altman is just saying, or for those of you who are paying attention to the wider AI community, something that Ilya Sutskov is saying, we have to also take a look at the wider AI community because Logan Kilpatrick, the person at Google AI, is actually saying that a straight shot to ASI is looking more and more probable by the end of the month. This is what Elya saw. And remember, he says the success of scaling test, time, compute, which is what Elia saw early signs of, is a good indication that this direct path is just continuing to scale up, might actually work up. So this is something that is truly fascinating because if you watch the talk from Elia Satskova about superintelligence, he actually gives some key details on where we might head next. Now, I just, you know, I'm astounded by this talk because it's just super insightful and I'm surprised we're getting all of this information from these key industry leaders. And of course his company is something that is super, super secret at the moment. But he does give us a recent update on where he thinks things will.

Head our incredible language models and they are unbelievable chatbots and they can even do things, but they're also kind of strangely unreliable and they get confused while also having dramatically superhuman performance on evals. So it's really unclear how to reconcile this. But eventually, sooner or later, the following will be achieved. Those systems are actually going to be agentic in a real ways, whereas right now the systems are not agents in any meaningful sense. Just very. That might be too strong. They're very, very slightly agentic. Just beginning. It will actually reason. And by the way, I want to mention something about reasoning. Is that a system that reasons? The more it reasons, the more unpredictable it becomes. The more it reasons, the more unpredictable it becomes. All the deep learning that we've been used to is very predictable because if you've been working on replicating human intuition, essentially it's like the gut feel, if you come Back to the 0.1 second reaction time, what kind of processing we do in our brains. Well, it's our intuition, so we've endowed our AIs with some of that intuition. But reasoning, you're seeing some early signs of that reasoning, is unpredictable. And one reason to see that is because the chess AIs, the really good ones, are unpredictable to the best human chess players. So we will have to be dealing with AI systems that are incredibly unpredictable. They will understand things from limited data, they will not get confused. All the things which are really big limitations. I'm not saying how, by the way, and I'm not saying when. I'm saying that it will and when all those things will happen together with self awareness, because why not self awareness is useful. It is part your ourselves are parts of our own world models. When all those things come together, we will have systems of radically different qualities and properties that exist today. And of course they will have incredible and amazing capabilities. But the kind of issues that come up with systems like this, and I'll just leave it as an exercise just to imagine it's very different from what we are used to. And I would say that it's definitely also impossible to predict the future, really, all kinds of stuff is possible. So.

So that was a snippet from a longer 20 minute talk where you can see that he basically says super intelligence is going to be agentic. It's going to be able to reason and it's going to be able to understand and be self aware. Now, if you're wondering how super intelligence actually looks when it arrives, actually found a very interesting snippet from this EXOP researcher that gives us an insight to how the next 3,500 days might actually look. And this is something that's truly fascinating because we don't really get insights from Exopi employees, but this is one that we should really pay attention to.

I personally would think that some combination of bigger models with more compute and models that have been more specifically trained to operate in these types of environments will succeed. And the question is just how long will it take to succeed? I think you could totally succeed in the next 12 months, but probably it will take a few more years than that because everything always takes longer than you expect. That's sort of one way of summarizing my view. I would imagine that if you could go inside these companies and look at their optimistic timelines, that it would be something like within 12 months we will have something like Claude 3.5 Sonic computer using Agent, except it actually works really well and we can just delegate tasks to it and have it running autonomously in the background doing all sorts of useful things for us. I would bet that there's like having a roadmap to try to get that this year, but realistically things take longer than you expect and there's going to be unforeseen difficulties and so forth. So where does the 2027 number come from? It's a combination of a bunch of different heuristics and a bunch of different trends and guesses and so forth. One thing I would say is that if you just do the obvious and very good thing of taking various benchmarks and extrapolating performance on those benchmarks, some of them get to superhuman performance this year, some of them get to superhuman performance next year. But like around 2027 in my subjective guess, based on all the benchmark extrapolations I've seen, is when it feels like I can say all the current benchmarks will be saturated.

So with that being said, what do you guys think about the future of superintelligence? Does it now seem closer that we've had many reputable industry figures actually talk about this being a reality? Or does it seem like another pipe dream to where we are all just in our hype bubble? And of course you can see here we have Gary Marcus A notable AI critic. This is going to be really fascinating because of course he is someone that I think is very important to the AI industry because he helps keep everyone in check. And he's basically saying here, will AI be able to, you know, do these eight things by the end of 2027? Will be able to, you know, watch an unseen mainstream movie without reading the reviews and be able to follow the plot twists, know when to laugh, basically testing the long form coherence of the model. I think that's going to be something that's really easy considering we already have needle in the haystack tests. So I'm not sure why that is. And he says, similar to the above, be able to read new mainstream novels without reading review views. Reliably answer questions about plot, character conflicts and motivations. Considering infinite memory and context. Windows are getting larger. I think it's going to be super simple. Write engaging brief biographies and obituaries without obvious hallucinations, that one seems pretty easy. But if we actually just gloss over this really quickly because it's kind of boring and we actually take a look at the last ones here that are apparently the most difficult, but with no little or human involvement, write Pulitzer caliber books of fiction and nonfiction, and that's going to be pretty difficult. And with little or no human involvement, write Oscar caliber screenplays. I think if you do have a model that's able to have some kind of framework where, let's say, for example, it generates 10,000 screenplays and then you have an audience of 20,000 and then you have that audience view them and then maybe you take the highest rated one, so maybe test time, compute for that, I think that could work. I know that sounds like a crazy framework, but if you had like an AI agent audience that was very similar to the audiences of today, maybe you just mapped out what the audience liked or whatever, I guess that could be something that could happen. And this one right here, with little or no human involvement, come up with paradigm shifting, noble caliber scientific discoveries. I think that is pretty insane because that would take a remarkable feat for AI to be able to do that. Of course, take proofs from the mathematical literature written in natural language and convert them into a symbolic form suitable for symbolic verification. And this is what he says in his newsletter. So of course Gary is a critic, so he's going to state this stuff and it'll be interesting to come back to 2027 to see if this stuff is, you know, done or not. But at the end of the day, I want to pose a question to you. Do you think this is going to be done? Do you think this is not going to be done? And I will see you guys in the next video.