This very thread is about LLMs hallucinating too much today. The curl dev is reporting a zero percent hit rate on automated bug reports in TYOOL 2025. You could say that's an example of the toupee fallacy, but nobody is showing up to defend the AI tools they used for valid curl bug reports either.

I'm still waiting to see an LLM that is actually helpful versus the alternatives, outside a very small set of tasks that's basically the same as "2 years ago".

They still hallucinate (generate bullshit) too much. Nothing much to be done there, it is part of what they are.

They are still useful as code assistant. But the sort of overhype that you push for is actually detrimental to its healthy development.