The era of boundary-breaking advancements is over? [video]
youtube.com
youtube.com
(Edit: plus a question mark, as we sometimes do with contentious titles.)
Maybe for LLMs but they are not the only possible algorithm. Only this week we had Genie 3 as in:
>The Surprising Leap in AI: How Genie 3’s World Model Redefines Synthetic Reality https://www.msn.com/en-us/news/technology/the-surprising-lea...
and:
>DeepMind thinks its new Genie 3 world model presents a stepping stone toward AGI https://techcrunch.com/2025/08/05/deepmind-thinks-genie-3-wo...
But are they sufficiently different that stalling progress in one doesn't imply stalling progress in the other?
Depends if you’re asking about real world models or synthetic AI world models.
One of them only exists in species with a long evolutionary history of survivorship (and death) over generations living in the world being modeled.
There’s a sense of “what it’s like to be” a thing. That’s still a big question mark in my mind, whether AI will ever have any sense of what it’s like to be human, any more than humans know what it’s like to be a bat or a dolphin.
You know what it’s like for the cool breeze to blow across your face on a nice day. You could try explaining that to a dolphin, assuming we can communicate one day, but they won’t know what it’s like from any amount of words. That seems like something in the area of neuralink or similar.
>It is comprised of a spatiotemporal video tokenizer, an autoregressive dynamics model, and a simple and scalable latent action model.
my point is more people can try different models and algorithms rather than having to stick to LLMs.
On the other hand, if you're telling your investors that AGI is about two years away, then you can only do that for a few years. Rumor has it that such claims were made? Hopefully no big investors actually believed that.
The real question to be asking is, based on current applications of LLMs, can one pay for the hardware to sustain it? The comparison to smartphones is apt; by the time we got to the "Samsung Galaxy" phase, where only incremental improvements were coming, the industry was making a profit on each phone sold. Are any of the big LLMs actually profitable yet? And if they are, do they have any way to keep the DeepSeeks of the world from taking it away?
What happens if you built your business on a service that turns out to be hugely expensive to run and not profitable?
Musk has been doing this with autonomous driving since 2015. Machine learning has enough hype surrounding it that you have to embellish to keep up with every other company's ridiculous claims.
Whether there is hype or not, the laws of money remain the same. If you invest and don’t get expected returns, you will be eventually concerned and will do something about it.
For me an AGI would mean truly at least human level as in "this clearly has a consciousness paired with knowledge", a.k.a. a person. In that case, what do the investors expect? Some sort of slave market of virtual people to exploit?
How to find out if something has probably consciousness? Much less clearly? What is consciousness?
LLMs are great at forming models of language from observations of language and extrapolating language constructs from them. But to get general intelligence we're going to have to let an AI build their models from direct measurements of reality.
They really aren't even great at forming models of language. They are a single model of language. They don't build models, much less use those models. See, for example, ARC-AGI 1 and 2. They only performed ARC 1 decently [0] with additional training, and are failing miserably on ARC 2. That's not even getting to ARC 3.
[0] https://arcprize.org/blog/oai-o3-pub-breakthrough
> Note on "tuned": OpenAI shared they trained the o3 we tested on 75% of the Public Training set. They have not shared more details. We have not yet tested the ARC-untrained model to understand how much of the performance is due to ARC-AGI data.
... Clearly not able to reason about the problems without additional training. And no indication that the additional training didn't include some feature extraction, scaffolding, RLHF, etc created by human intelligence. Impressive that fine tuning can get >85%, but it's still additional human directed training and not self contained intelligence at the level of performance reported. The blog was very generous making the undefined "fine tuning" a footnote and praising the results as if they were directly from the model that would have cost > $65,000 to run.
Edit: to be clear, I understand LLMs are a huge leap forward in AI research and possibly the first models that can provide useful results across multiple domains without being retrained. But they're still not creating their own models, even of language.
Think about this story https://news.ycombinator.com/item?id=44845442
Med-Gemini is clearly intelligent, but equally clearly it is an inhuman intelligence with different failure modes from human intelligence.
If we say Med-Gemini is not intelligent, we will end up having to concede that actually it is intelligent. And the danger of this concession is that we will under-estimate how different it is from human intelligence and then get caught out by inhuman failures.
I guess when it comes to the definition of intelligence, just like porn, different people have different levels of tolerance.
I believe that’s Eliezer Yudkowsky’s definition.
It doesn't matter whether they are lying. People want to hear it. It's comforting. So the market fills the void, and people get views and money for saying it.
Don't use the fact that people are saying it, as evidence that it is true.
It's not the AGI sceptics who are getting $500bn valuations.
You mean like euclidean geometry?
>> obvious mathematical certainty
> You mean like euclidean geometry
To which you reply stuff about values and bubble bursts.
What you said may or may not be true. But it's hard for me to tell how it is related to what I asked.
So I wanted to understand what you mean by this concept.
But we can leave it. It's not important. No harm meant.
In other words, the statement "I know I am not chasing a bubble because valuations are high and growing at an increasing rate" is categorically false because this property is a common indicator of the exact opposite; that you are in an exponentially growing bubble.
The only thing I said is: there are probably no _obvious_ mathematical certainties in reply to what you said, as the history of mathematics shows with plenty of examples.
But it's already too much ink spilled for such a little difference.
I reckon now that you use words like "obvious" (or "categorically" in the last reply [1]) as markers of belief strength or something. That's fine, many people do it.
[1] related: https://news.ycombinator.com/item?id=36808333
The only thing I said is: there are probably no _obvious_ mathematical certainties in reply to what you said, as the history of mathematics shows with plenty of examples."
There is nothing more obvious and certain mathematically than the fact that exponential growth of bubbles entails increasing valuations and increasing rates of growth in those valuations.
I use the word "obvious" quite properly as in "can be read from any graph", or "is contained within the structure or the model". I use categorical because the implied conclusion of the original comment is a categorical error.
https://math.stackexchange.com/questions/820686/obvious-theo...
I think you conflate two questions in your statement: the question of what can be said about a function that is up to a point exponential and growing, and after that point decaying in an unspecified manner, but so that this point is a local maximum. And secondly the question if this is a good/the only/the best model for valuations in what is called an investment bubble that then bursts.
This conflation of model and modeled makes it hard for me to assess the argument, because I need to take these layers apart, and that involves guess work.
> The greatest values, and change in values comes just before the bubble bursts.
This seems to be talking about the mathematical model. Then you go on:
> In fact, tulips were most valuable, and those values were climbing at the highest rate before the tulip bubble burst.
That seems a jump to the modeled. And you speak of the modeled as if it was the model.
And by the way, if this graph [1] is correct, then what you say is also false. The valuations do not rise exponentially, but slow before the bubble bursts, very visibly. This is also true for the 2008 housing bubble in the local index [2].
I think that shows quite well that there are two layers. And there is the question of the fit between them. Having seen those two graphs I think I'd rather expect reduction in the rate of change of valuation shortly before the bubble bursts. That model appears to fit better what I'd expect in economical bubbles (but not in soap bubbles), namely a growing sense of risk that counteracts the fomo.
[1] image Tulip Price Index 1636-37 here https://www.historydefined.net/how-did-tulips-create-the-wor... [2] https://wolfstreet.com/wp-content/uploads/2021/07/US-Housing...
There are interesting and well thought out arguments for why the AGI is not coming with the current state of technology, dismissing those arguments as propaganda/clickbait is not warranted. Yannic is also an AI professional and expert, not one to be offhandedly dismissed because you don’t like the messaging.
Telling us all to remember that there's potential for bias isn't so bad. It's a hot button issue.
There was no grand announcement of passing the Turing test or not. Instead the whole idea has faded in importance.
As the models get better, it wouldn't be shocking to me we get to a point that no one cares if the models are considered "AGI" or not.
We will be chasing some new vaguely defined concept.
Imagine we trained an AI on everything ever written, but the catch is we’ve restricted the training data to the year, let’s say, 400 BCE and earlier (Ignore the fact that most of what was written then is lost to us now and just pretend that’s not issue for our thought experiment) The AI is also programmed to seek new knowledge based off that starting knowledge.
Also pretend that this AI has an oracle it could talk to that would help the AI simulate experiments. So the AI could ask questions and get answers but only in a way that builds ever so slightly off what it already knows.
Making any progress at all in this experiment and discovering new knowledge is what we’re after and “new knowledge” would be defined as some r that’s demonstrated using some p and q as propositions, where r is neither p or q, and r is also correct in terms of 2025 knowledge.
If the AI, with the aid of the knowledge it started with and the help of the oracle to let it ask questions about the world and build off that knowledge, can ever arrive, or exceed, 2025 knowledge then it’s at least generally intelligent and equal to a human. Although the bar could maybe be even less.
It loses, however, if it never advances, gets stuck in a loop, or in some other sense can’t make progress.
This is intelligence: to proceed from things everyone agrees on and ask questions and formulate assertions that depend on propositions holding true, and in the process demonstrate new things that were not already part of common belief.
I don’t know how this experiment could be done in real life with an LLM for example but this is a story version of what I mean in my original comment.
I tried this, and screamed "Goal!!!!" on a golf course when the ball went in the hole. Did not work as expected. Lost my membership.
But maybe I applied it wrong.
The default position that does not need any more justification is the one that is skeptic or even agnostic to the claim that is made until proof is shown.
So when talking about evidence as a way to prove a claim: AGI is coming is the team that needs to provide this evidence. Someone saying AGI is not coming can add as many arguments or opinions as they like but it does not usually invite to such a high scrutiny as saying they need to provide evidence.
"Modern" definitions that include non-intelligence related stuff like agency sound like goalpost moving, so it's unclear why would you want them.
Your case with locked-in syndrome is interesting. Can human brain develop intelligence in full absence of any senses? I doubt so.
You can make it stronger at being cross domain, but it satisfies the minimum requirement.
No, it is not. It is trained to predict next token, and it is trained to follow user instructions.
I still disagree though, they are not trained to practice law. They are trained to remember law, but practice come from general training to follow instructions.
Given that AGI does not exist, “AGI is not coming” is the status quo until someone disproves it.
The inverse can be true too: Just because people ARE saying that Agi is coming, isn’t evidence that it is true.
"AI is getting better rapidly" is the current state of affairs. Arguing "AI is about to stop getting better" is the argument that requires strong evidence.
Llms getting better != a path to AGI.
… is it? I hear people saying that. I see “improvement”: the art generally has the right number of fingers more often, the text looks like text, the code agents don’t write stuff that even the linter says is wrong.
But I still see the wrong number of fingers sometimes. I still see the chat bots count the wrong number of letters in a word. I still see agents invent libraries that don’t exist.
I don’t know what “rapid” is supposed to mean here. It feels like Achilles and the Tortoise and also has the energy costs of a nation-state.
You've frog-boiled yourself into timelines where "No WORLD SHAKING AI launches in the past 4 months" means "AI is frozen". In 4 months, you will be shocked if AI doesn't have a major improvement every 2 months. In 6 months, you will be shocked if it doesn't have a major update ever 1 month.
It's hard to see exponential curves while you're on it, I'm not trying to fault you here. But it's really important to stretch yourself to try.
There's been the obvious notion that digitizing the world's information is not enough and that hasn't changed.
You assume everyone is “impressed”.