@arstechnica Ok, ChatGPT is bad.

It's bad in an absolute sense and relative to almost anything else in the LLM space.

Literally, if I start a Qwen3-Next 80B (or the next bigger one) locally on my PC without any grounding, I'll probably get quality-wise similar answers.

So why is it still in use?

It's not an “intelligent” system in any way; it's a statistical human text parrot.
Don't take me wrong; you can build useful, somewhat “intelligent” systems on LLMs, but ChatGPT isn't it.