Back in 2001, our social medium was Slashdot and no one ever pretended to read the article. No one read the article either. It was slashdotted most of the time anyways.
State-sponsored psyop meta comments aside, the models obviously continue to get better, but there is still a lot of 'guard railing' required to keep even the latest models completely on-task. The chess example is interesting because it's clearly a well-studied and established domain so the rules, strategies, and whatever else is in the training data should make yield excellent results; but clearly there is some behavior in these systems that's difficult to engineer out.
I'm not sure why anyone is expecting stochastic systems to be deterministic.
Chess is a deterministic game won by a combination of known movesets and constrained multi-level forward search.
LLMs do neither of these things. They don't reproduce training data exactly, their next response is more 'inspired by' prompts and its own memory than produced deterministically, and they don't have the capability to do general forward search on their own.
So when you ask an LLM to play chess you're getting the equivalent of a very compressed and lossy JPEG of chess rules and strategies with added per-turn random noise.
They also don't have the ability to design their own chess engine, although it would be interesting to see what happens if you ask for one.
For me the useful intuition is that LLMs haven't somehow magickally learned to implement any of the algorithms we know that we have used to make strong chess engines: alpha-beta minimax and Monte-Carlo Tree Search on the one hand, and obviously the ability to learn accurate evaluation functions by self-play.
I mean we've done all this before in a task-specific fashion. It's useful to know that LLMs haven't managed to do that in the process of learning to represent the entire text on the web. On the other hand they have gotten say very good at machine translation without being trained exclusively (and I select the preceding word carefully) on machine translation.
Edit: I'm saying this because there is this idea expressed by e.g. Ilya Sutskever, that in order to predict the next token accurately an LLM has to learn something about all of underlying reality. See for example this interview with Dwarkesh:
Where Sutskever claims that "Predicting the next token well means you understand the underlying reality that led to the creation of that token".
If that were true, we should have seen LLMs play good chess by now. There is a huge amount of data on playing chess floating around on the web in the form of algebraic chess notation and if LLMs were capable of learning the "underlying reality" of chess, they would already have. They haven't. Because they can't. What Sutskever is saying flies in the face of literally hundreds of years of statistical modelling, which is to say, building predictive models that, very explicitly, do not have to understand any "underlying reality" and only have to be good at modelling a dataset.
>If that were true, we should have seen LLMs play good chess by now.
Not at all. LLMs learn by imbibing a mass of relationships as isolated fragments of information. There is a certain amount of sorting and indexing that happens during the training phase. There is also a certain amount of compute executed on these relationships during inference. LLMs can model processes that fit within the compute budget. Language translation works well because language is lookup-heavy while being light on compute.
Chess is a compute heavy game of finding the best move out of many possibilities with wide variation in the quality of each move. Humans cut through the compute requirements by reinforcement and learning intuition. LLMs don't get reinforcement on chess so they must compute during inference a unified model of chess. Developing a strong model of chess from raw fragments of information is simply not in their compute budget.
>> Not at all. LLMs learn by imbibing a mass of relationships as isolated fragments of information.
You gotta be careful how you use the word "relation" here because there's an informal meaning (I'm related to my cousin) and a more strict, formal meaning, that is used in computer science e.g. in the "Relational Calculus" etc. In the formal sense, the one relation that LLMs learn during training is the co-occurrence of tokens in a corpus of text, what's called more technically a "collocation" relation. Nothing says that this is enough to play chess, so I'm indeed doubtful that they can.
But they have "learned to implement any of the algorithms we know that we have used to make strong chess engines". Ask Claude Code to write you a chess engine. Your objection is that they don't implement MCTS in the neurons themselves? Neither does a human, we use a C compiler when we want to play chess using MCTS.
That's a separate question from whether an LLM (unaided by a C complier) can learn to play chess as well as human (also unaided by a C compiler). Certainly humans can't become grandmasters only by reading chess transcripts on the web, and certainly humans require many "thinking tokens" during a game to play effectively. Do you know for sure that a transformer can't reach grandmaster level if it is allowed to learn by playing games (as humans do) and is given a sufficient number of thinking tokens during the game? It seems near certain that they could, if someone wanted to spend the money (and I don't see why anyone would.)
Sutskever's claim is that in order to predict the next token a system must learn something about the "underying reality" that produced the token. In the context of chess that means that the LLM must learn something about playing chess (since tokens are the moves in a game of chess). My argument is that contrary to what should be expected if we take what Sutskever says to be true, they don't seem to have.
Yes, I do mean that the LLM's weights are set so that it will execute minimax or MCTS when it needs to. That has nothing to do with whether humans can do the same or not.
I don't disagree that a Transformer could learn to play chess if it was explicitly trained to do that. My argument is that LLMs, trained to predict the next token, have not learned to play chess. That's LLMs, not Transformers.
Just to make sure this is not taken as splitting hairs, the point is that there's all sorts of claims made about what LLMs learn when they train on text. For example, there was a claim by Sundar Pichai that one of their models had learned to translate Bengali without explicitly being trained to do so. It later emerged that Bengali was indeed included in the model's training set [1]. It's not clear whether that included parallel texts, e.g. between Begnali and English or another intermediary language, in any case Sundar Pichai's claim was that the ability to translate Bengali was "emergent".
So I'm interested in understanding the extent to which these "emergent" abilities are real or not. With chess, given the amount of textual data tracing games that floats about on the open internet, I would totally except some ability to play chess to "emerge". Maybe the reported 700-800 ELO level is even that sort of ability. Maybe we should only expect LLMs to learn to play at the level of an untrained, casual player. Maybe not. I have no idea.
On the other hand, the fact they keep making elementary mistakes like illegal moves must be taken to mean that, so far, LLMs haven't learned to play chess.
But it speaks in words, therefore it must be super duper extra smart!!11 /s
Sarcasm aside, I think this is an easy cognitive trap to fall into. It does sometimes feel like the LLM must have some world model because it converses somewhat coherently. Examples like this failure to understand chess, or to count the number of Rs in "strawberry", seem difficult to explain if the models are intelligent. But that doesn't stop people believing they are anyway. I think there must be something about the conversational interface that fools us easily. I wonder if people trained in interrogation techniques are also fooled?
Humans are a bunch of individuals though with an incredible amount of diversity. If their claim is biological, it would be helpful to pin down more precisely what they mean, because right now it's pretty vague. I'm dubious that there's a rigorous claim here though because pretty much every time I've seen a claim like this before it's essentially just protection of one's personal beliefs about psychology without more of a basis than assumption that their personal experience is universal, which is what the comment you're responding to is alluding to.
They didn't explicitly say they were trying to speak for all of humanity. I'm responding to the specific interpretation of it directly because as the comment you were responding to pointed out, it was ambiguous.
In typical HN fashion, the blog post is a run of the mill “hack” and instead of commenting on the article, a Hacker News instead word vomits their favorite IKEA furniture.
Guidelines explicitly mention this type of submissions:
> Off-Topic: Most stories about politics, or crime, or sports, or celebrities, unless they're evidence of some interesting new phenomenon. If they'd cover it on TV news, it's probably off-topic.
You think it is an interesting submission, but it isn't even a particularly relevant step in the ongoing soccer drama.
Calling Swift middle class and that they were possibly paycheck to paycheck is laughable.
Her father had a huge role to play in her success with his business deals and would largely be considered upper middle class if not higher. If anything, Taylor’s story reinforces the idea that families with wealth offer easier paths in life for their children.
I'm from Pennsylvania and a proverbial stone's throw from where Taylor Swift is also from. Trust me when I tell you that there's nothing "middle class" about her. Her parents are wealthy and this gave her a big boost into the entertainment industry, with lots of expensive coaching, music lessons, and trips to Nashville.
IIRC he worked for a large financial services company (Merrill Lynch?) and was able to transfer to the Nashville office so that Taylor could be closer to the music scene.
No offense, but "the idea that families with wealth offer easier paths in life for their children" - like isn't this obviously the case?
The amazing thing about America is that there are paths for those not as wealthy, but of course they are harder, as much as we might not want them to be... no?
The thread is about Taylor Swift, arguably the most successful music entertainer of all time. The parents with upper|middle class gave a boost, but you absolutely cannot buy this level of stardom and hasn't been replicated anywhere that I know of.
America is 100% the place to be rich or poor, if this really is your goal, from Dolly to Jackson or Swift.
Nobody is claiming this or ever should. The claim is that there is real proof of "success" from dirt poor to middle|upper class with these music entertainer examples. Nothing ever will be equal chance, and that shouldn't really be the goal [depending on how it's defined and implemented].
reply