Things like "What is today's date" used to be enough (would usually return the date that the model was trained).
I recently did things like current events, but LLMs that can search the internet can do those now. i.e. Is the pope alive or dead?
Nowadays, multi-step reasoning is the key, but the Chinese LLM (I forget the name of it) can do that pretty well. Multi-step reasoning is much better at doing algebra or simple math, so questions like "what is bigger, 5.11 or 5.5?"
I recently did things like current events, but LLMs that can search the internet can do those now. i.e. Is the pope alive or dead?
Nowadays, multi-step reasoning is the key, but the Chinese LLM (I forget the name of it) can do that pretty well. Multi-step reasoning is much better at doing algebra or simple math, so questions like "what is bigger, 5.11 or 5.5?"