The best framework I can use to explain it is that these LLM are acting a role.
They are predicting the next token out of an almost infinite depth of consciousness. They fit in in what happens from the previous tokens.
They can play the idiot really well if you make them think this is the role.
So do not insult them for doing a poor job, they are not humans. They will not find in the depth of their pride new abilities, they will simply start playing the role of the human that does a poor job. Fitting in the role perfectly.
顯示更多
When ChatGPT detects your IQ is under 90 it starts spamming emojis