Ilya Finally Bends the Knee: "DeepMind Was Right!”

This interview marks the start of a new chapter in AI. The moment when Ilot Saskcover, the last man standing, finally bent the knee and admitted that Deis and Deep Mind had it right all

along. They believed that we need a

along. They believed that we need a couple of breakthroughs that aren't even theorized properly. Essentially, we

theorized properly. Essentially, we don't know what we are missing, but it is important. Something strange is going

is important. Something strange is going on. One of the very confusing things

on. One of the very confusing things about the models right now. How to

reconcile the fact that they are doing so well on >> eval >> and you look at the evals and you go those are pretty hard evals.

>> The economic impact seems to be dramatically behind and it's like how is that possible?

>> Yeah.

>> It's like I'm not sure. Before we begin though, it's important to remind you who I is and how his stance can move the entire AI landscape. Jeff Hinton is

known as the pioneer who invented or at least popularized many of the core ideas behind neuronet networks and Ilia as his student became one of the first people

on the planet standing at the crossroads of those ideas and the compute power needed to actually make them real and that made Iliot a very special person.

In 2015, when Elon Musk was laying the foundations of OpenAI, he had to fight for Ilia, pulling him away from Google.

And Ilia went on to become one of Open Eai's founding members and its first chief AI scientist. Then he led the research all the way to Project Strawberry or QSAR, which came out as

GPT01 or the first thinking model ever created. The important thing is although

created. The important thing is although Google was claiming they had similar stuff to Chad GPT internally, the reality was GP3 and GPD4 were miles ahead. Even Google was surprised. So

ahead. Even Google was surprised. So

Ilia is not just a former Open AI scientist. He is the person who put

scientist. He is the person who put language models on the map. But when

Ilia founded SSI a couple of years back, he posted mountain identified time to climb. Referring to his belief that we

climb. Referring to his belief that we have everything we need, it's just a matter of building it. which goes

against the stance of Deis Hassabes, Yan Lon and Frano Chile which altogether make the key figures who effectively set the intellectual direction of modern AI.

They believed that we need a couple of breakthroughs that aren't even theorized properly. Essentially, we don't know

properly. Essentially, we don't know what we are missing, but it is important. Now, after a year of focused

important. Now, after a year of focused research, Ilia finally broke his stance, effectively ending the entire debate and uniting the field. surprised by how good

are the models on evaluations but not in economic output.

>> They are doing so well on evals.

>> Mhm.

>> And you look at the evals and you go those are pretty hard evals.

>> The economic impact seems to be dramatically behind. I have two possible

dramatically behind. I have two possible explanations. Maybe a real training

explanations. Maybe a real training makes the models a little bit too single-minded and narrowly focused and because of this they can't do basic things. But there is another explanation

things. But there is another explanation which is back when people were doing pre-training the question of what data to train on was answered because the

that answer was everything.

>> Yeah.

>> So you don't have to think is it going to be this data or that data.

>> Yeah.

>> But when people do RL training they do need to think. They say okay we want to have this kind of RL training for this thing and that kind of training for that thing. And from what I hear, all the

thing. And from what I hear, all the companies have teams that just produce new RL environments and just add it to the training mix. And then the question is, well, what are those? There are so many degrees of freedom. There is such a

huge variety of RL environments you could produce. One thing you could do,

could produce. One thing you could do, and I think that's something that is done inadvertently, is that people take inspiration from the evals. You say,

"Hey, I would love our model to do really well when we release it. I want

the evals to look great." If you combine this with generalization of the models actually being inadequate, that has the potential to explain a lot of what we

are seeing. This disconnect between eval

are seeing. This disconnect between eval performance and actual real real world performance, which is something that we don't today exactly even understand what

what we mean by that. language math and coding and especially math and coding suggests that whatever it is that makes people good at learning is probably not so much a complicated prior but

something more some fundamental thing.

>> Wait, I'm not sure I understood. Why

should that be the case?

>> So consider a skill that people exhibit some kind of great reliability. If the

skill is one that was very useful to our ancestors for many millions of years, hundreds of millions of years, you could argue that maybe humans are good at it

because of evolution because we have a prior, >> an evolutionary prior that's encoded in some very nonobvious way.

>> Yeah.

>> That somehow makes us so good at it.

>> Yeah. But if people exhibit great ability reliability robustness ability to learn in a domain that really

did not exist until recently, then this is more an indication that people might have just better machine learning period. So now all of the most

period. So now all of the most influential AI scientists have agreed on two seemingly contradictory points. one

to achieve AGI we need something different a sort of breakthrough that isn't even theorized at the moment but interestingly that doesn't mean the current AI path is hitting a wall the

current AI paradigm and scaling laws will continue to deliver as Ilia doubled down on X scaling the current thing will keep leading to improvements in

particular it won't stall but something important will continue to be missing the thing that made this interview very special is that Ilia went on to describe

some of the characteristics he expects a true AGI would have. If you think about the term AGI, you will realize and especially in the context of pre-training, you will realize that a

human being is not an AGI because a human being, yes, there is definitely a foundation of skills. A human being lacks a huge amount of knowledge.

Instead, we rely on continual learning.

And so then when you think about okay so let's suppose that we achieve success and we produce a safe super some kind of safe super intelligence the question is but how do you define it where on the

curve of continual learning is it going to be I produce like um a super intelligent 15year-old that's very eager to go and you say okay I'm going to they don't know very much at all the great

student very eager you go and be a programmer you go and be a doctor go and learn so you could imagine that the deployment itself will involve some kind

of a learning trial and error period.

It's a process as opposed to you drop the finished thing.

>> Okay. I I I I see. So you're you're suggesting that the thing you're pointing out with super intelligence is not some finished

mind which knows how to do every single job in the economy cuz the way say the original I think open AAI charter or whatever defines AGI is like it can do every single job that a every single

thing a human can do. You're proposing

instead a mind which can learn to do any single every single job.

>> Yes.

>> And that is super intelligence. And then

but once you have the learning algorithm it gets deployed into the world the same way a human laborer might join an organization.

>> So he expects a learning machine not a finished oracle machine that generalizes better and gathers experience on the job that needs a new machine learning

paradigm. And Ilia said I have some

paradigm. And Ilia said I have some ideas. How can you know the teenage

ideas. How can you know the teenage driver kind of self-correct and learn from their experience without an external teacher? And the answer is well

external teacher? And the answer is well they have their value function >> right? They have a general sense which

>> right? They have a general sense which is also by the way extremely robust in people like whatever it is the human value function whatever the human value function is

with a few exceptions around addiction it's actually very very robust. And so

for something like a teenager that's learning to drive, they start to drive and they already have a sense of how they're driving immediately. How badly

they're unconfident and then they see okay and they and then of course the the learning speed of any teenager is so fast after 10 hours you're good to go.

>> Yeah, it seems like humans have some solution. But I'm curious about like

solution. But I'm curious about like well how are they doing it and like why is it so hard to like how do we need to reconceptualize the way we're training models to make something like this possible? You know that is a great

possible? You know that is a great question to ask and it's a question I have a lot of opinions about but unfortunately we live in a world where not not all

machine learning ideas are discussed freely and this is this is one of them.

So there's probably a way to do it.

I think it can be done. The fact that people are like that I think it's a proof that it can be done. But

regardless, I do think it points to the existence of some machine learning principle that I have opinions on, but unfortunately circumstances make it hard

to to discuss in detail. So far, the interview might sound a little bearish, but it was the complete opposite because Ilia believes that we haven't seen the real power of AI yet.

>> From the average person's point of view, nothing is that different will continue being true even into the singularity.

No, I don't think so. Okay. So, such and such company announced some difficult to comprehend dollar amount of investment, >> right?

>> I don't think anyone knows what to do with that.

>> Yeah.

>> But I think that the impact of AI is going to be felt. AI is going to be diffused through the economy. There are

very strong economic forces for this and I think the impact is going to be felt very strongly and we'll see. I think

that the world will truly change as AI becomes more powerful.

>> Yeah. And I think a lot of these forecasts will like I think things will be really different and people will be acting really differently.

>> What speaking of forecasts what are your forecast to this system you're describing which can learn as well as a human and subsequently as a result becomes

superhuman.

>> I think like uh 5 to 20 >> 5 to 20 years.

>> Mhm.

>> The part that surprised me was his take on compute. He said historically

on compute. He said historically breakthroughs didn't need massive compute. Of course, they can be scaled

compute. Of course, they can be scaled up, but they show promise on a smaller scale. So even at this point, this is

scale. So even at this point, this is still anyone's game. If the compute needs aren't forbidding, then we might wake up to a new paradigm from a small lab in Singapore someday. I remember

when Chad GPT first came out, I was listening to a researcher. I can't

recall his name now, but he said something that stuck with me. Chad GPT

is going to delay AGI, not bring it closer. All the major labs, especially

closer. All the major labs, especially OpenAI, stopped publishing their best work and the scientific process at the frontier basically collapsed. Instead of

3 years of shared collaborative progress, everyone ended up rebuilding the same thing in 10 different places.

Every major lab is racing toward AGI on its own, a duplication of effort that undoubtedly slows progress. At the end, Ilia talked about SSI, how he might adjust his plans on a straight shot at

super intelligence, how he's going to make money, and why SSI has a real shot at winning the race. There are some ideas that I think are promising, and I

want to investigate them and see if they are indeed promising or not. It's really

that simple. It's an attempt. I think

that if the ideas turn out to be correct, these ideas that we discussed around understanding generalization, >> if these ideas turn out to be correct, then I think we will have something

worthy.

>> How will SSI make money?

>> You know, my answer to this question is something like we just f right now we just focus on the research and then the answer to that question will reveal itself. I think

there will be lots of possible answers.

M is SSI's plan still to straightshot super intelligence?

>> Maybe.

I think that there is merit to it.

>> I think there's a lot of merit because I think that it's very nice to not be affected by the day-to-day market competition. But I think there are two

competition. But I think there are two reasons that may cause us to change the plan. one is pragmatic if timelines turn

plan. one is pragmatic if timelines turn out to be long and second I think there is a lot of value in the best and most

powerful AI being out there impacting the world >> competition like competition loves specialization and you see it in the market you see it in evolution as well so you're going to have lots of

different niches and you're going to have lots of different companies who are occupying different niches in in this kind of world where you might say Yeah, like one AI company is really quite a

bit better at some area of really complicated economic activity and a different company is better at another area and the third company is really good at litigation and that's contradicted by what humanlike learning

implies is that like it can learn.

>> It can but but you have accumulated learning. You have a big investment. You

learning. You have a big investment. You

spent a lot of compute to become really really really good really phenomenal at this thing and someone else spent a huge amount of comput and a huge amount of experience to get really really good at some other thing right

>> I think there is no argument left as all the major AI scientists are now on the same side AGI needs a couple of breakthroughs the current paradigm doesn't hit a wall but it continues to

miss something important and true AGI is something between 5 to 20 years away one thing that still remains a mystery is is that what would happen if we just

continue to scale the current paradigm.

If you just look at Gemini 3's emergent behaviors, it's really hard to imagine what would happen when we increase the compute by 100 times, which will happen in the next couple of years. Thanks for

watching. I see you in the next