lemonfever

·5 か月前·議論

Most LLMs output a whole bunch of tokens to help them reason through a problem, often called chain of thought, before giving the actual response. This has been shown to improve performance a lot but uses a lot of tokens