less than 1 minute read

There are many ways to waste one’s time. Debating about something that has a vague definition is one. When people argue against each other on questions like “is ChatGPT intelligent?”, it mostly goes like this: they suggest their own definition of intelligence and tell whether ChatGPT’s capabilities fulfill their requirements.

I think we should admit that the term is simply overloaded and not well-defined. In most debates like this, you can simply replace “intelligent” with “good” and see no difference—saying an LLM is intelligent is as vague as saying an LLM is good. The compression rate is so high that the sentence does not include much information beyond whether the person liked it or not.

I rather prefer to talk about whether a model A can do task B well based on a criterion C. That way, you lose no information. The claim can be proven or disproven by scientific evidence.

Updated: