Comment on It's not just statistics: GPT-4 does reasonparentComments−pixl973yLLMs don't see words like you do. Tokenization makes it behave odd. Often you can get 4 to output a solution in code and from that it derives a correct answer.
Comments
LLMs don't see words like you do. Tokenization makes it behave odd. Often you can get 4 to output a solution in code and from that it derives a correct answer.