4 ms·
I'm not convinced that what is learned from chess, or checkers for that matter, translate to any sort of general GPS or for example of understanding language.
by wglb 2y ago
I'm not convinced that what is learned from chess, or checkers for that matter, translate to any sort of general GPS or for example of understanding language.
Terry Winograd's https://en.wikipedia.org/wiki/SHRDLU https://en.wikipedia.org/wiki/SHRDLU doesn't seem to borrow ideas from either of those two fields, nor does it seem that modern day LLMS borrow from either of these three endeavors.
- QueensGambit 2y agoSure, LLMs are very different from chess algorithms. The endgame in chess is similar to deterministic algorithms in programming, while LLMs use probabilistic models. This is why the next AI improvement might come from combining deterministic and probabilistic models, similar to how chess requires different strategies for the endgame and the middle game.