Researchers asked frontier models to play chess. Have a look at the MAR rates in Table 3. When not explicitly told which moves were legal, no model identified legal moves at a rate better than 80%. Many asked for more illegal moves than legal moves. And even when explicitly told which moves were legal, the models continued to ask for illegal moves. With illegal asks discarded, none of the bots could beat a chess model calibrated to 1100 ELO.
The author of the originating post says that "current frontier models need laborious oversight and guardrails on even the simplest tasks", and he's absolutely correct.
It is even worse.. This is a classical reinforcement problem where data generation is easy because the rule set is pre-defined. So you really don't even need any data to start with (but would help).
There are more possible game combinations than atoms in the universe, even those generation of valid game states are as you say pre-defined. that is why models cannot go this route and therefore are poor at chess
Isn’t this exactly how AlphaZero was trained? The rules are known and well defined so the training process can generate games without any outside data.
The only reason LLMs are this bad at chess is because the labs don’t care about chess performance so they’re not going out of their way to train the models for it. The ability they do have is from what chess information happens to be in the training data, plus whatever general reasoning abilities they may be able to apply.
carodgers · · focus · HN ↗
<a href="https://arxiv.org/html/2509.24239v4" rel="nofollow">https://arxiv.org/html/2509.24239v4
Researchers asked frontier models to play chess. Have a look at the MAR rates in Table 3. When not explicitly told which moves were legal, no model identified legal moves at a rate better than 80%. Many asked for more illegal moves than legal moves. And even when explicitly told which moves were legal, the models continued to ask for illegal moves. With illegal asks discarded, none of the bots could beat a chess model calibrated to 1100 ELO.
The author of the originating post says that "current frontier models need laborious oversight and guardrails on even the simplest tasks", and he's absolutely correct.
threethirtytwo · · focus · HN ↗
The caveat is: It depends on the task.
Are there reams of chess moves that the model can train off of? No.
Are there reams of math papers the model can train off of? Yes.
tjwebbnorfolk · · focus · HN ↗
This is as false as something can possibly be. There are open databases of millions of chess games spanning hundreds of years.
XenophileJKO · · focus · HN ↗
manquer · · focus · HN ↗
wat10000 · · focus · HN ↗
The only reason LLMs are this bad at chess is because the labs don’t care about chess performance so they’re not going out of their way to train the models for it. The ability they do have is from what chess information happens to be in the training data, plus whatever general reasoning abilities they may be able to apply.