4 ms·
This seems quite likely to me, but did they special case it by reinforcement training it into the LLM (which would be extremely interesting in how they did it a
by dmurray 2y ago
This seems quite likely to me, but did they special case it by reinforcement training it into the LLM (which would be extremely interesting in how they did it and what its internal representation looks like) or is it just that when you make an API call to OpenAI, the machine on the other end is not just a zillion-parameter LLM but also runs an instance of Stockfish?
- shaky-carrousel 2y agoThat's easy to test, invent a new chess variant and see how the model does.
- gliptic 2y agoBoth an LLM and Stockfish would fail that test.
- delusional 2y agoNobody is claiming that Stockfish is learning generalizable concepts that can one day meaningfully replace people in value creating work.
- droopyEyelids 2y agoThe point was such a question could not be used to tell whether the llm was calling a chess engine
- delusional 2y agoAh okay, I missed that.
- andy_ppp 2y agoYou're imagining LLMs don't just regurgitate and recombine things they already know from things they have seen before. A new variant would not be in the dataset so would not be understood. In fact this is quite a good way to show LLMs are NOT thinking or understanding anything in the way we understand it.
- shaky-carrousel 2y agoYes, that's how you can really tell if the model is doing real thinking and not recombinating things. If it can correctly play a novel game, then it's doing more than that.
- jahnu 2y agoI wonder what the minimal amount of change qualifies as novel? "Chess but white and black swap their knights" for example?
- the_af 2y agoI wonder what would happen with a game that is mostly chess (or chess with truly minimal variations) but with all the names changed (pieces, moves, "check", etc, all changed). The algebraic notation is also replaced with something else so it cannot be pattern matched against the training data. Then you list the rules (which are mostly the same as chess). None of these changes are explained to the LLM, so if it can tell it's still chess, it must deduce this on its own. Would any LLM be able to play at a decent level?
- jahnu 2y agoNice. Even the tiniest rule, I strongly suspect, would throw off pattern matching. “Every second move, swap the name of the piece you move to the last piece you moved.”
- dwighttk 2y agoNo LLM model is doing any thinking.
- selestify 2y agoHow do you define thinking?
- 2y ago
- dmurray 2y agoIn both scenarios it would perform poorly on that. If the chess specialization was done through reinforcement learning, that's not going to transfer to your new variant, any more than access to Stockfish would help it.