Thanks, but Yis have to stop focusing on standardized test scores.
Yup, they could do Ruozhiba or Bilibili benchmark which is really chanllengeing and funny in the same time
@Yhyu13 That's interesting. I looked up those benchmarks. Apparently a Chinese humor forum proves to be an effective LLM test. Additionally, fine-tuning with the test generally boost the performance of LLMs.
This kinda makes sense because human geniuses reliably have a sarcastic wit, and none are moralizing censoring pinheads like modern LLMs. We should be adding humor to LLMs instead. It will help them to form self-consistent coherent world views based on cognition rather than 'because I said so' ideological nonsense like religion, moral relativism, and nationalism.
Anyways, Yi models have some nice abilities, including surprisingly good eloquence at times which includes the effective use of metaphors (rare with LLMs), but their lack of broad knowledge and ideological censorship is making them less than useless to educated free thinking intellectuals. Adding some wit should help.