I've come to cease all "inquiry" type usage of LLMs because of this. You really can't trust anything they say at all that isn't verified by a domain expert. But I can let it write code for me, and the proof is in the PR. I think ultimately the real value in these things is agentic usage, not knowledge generation.