Hallucination
also: hallucinations, hallucinated, hallucinating, hallucinate
When an AI confidently says something that isn't true, like inventing a function that doesn't exist or claiming tests passed when it never ran them.
Output from a language model that is fluent and confident but false or unsupported, such as nonexistent functions, package names, file paths, or settings, or false claims about actions it took. It happens because models generate plausible text rather than looking facts up.
A very confident tour guide who, when he doesn't know an answer, makes up a convincing one instead of saying "I don't know." It sounds exactly like his real facts.
Agents sound equally sure when they're right and when they're wrong. The fix is evidence: ask to see test output, error messages, documentation, or the actual file, rather than trusting a summary.
I may have hallucinated that option. I don't see retryOnFailure in the library's type definitions. Let me check the docs and use a supported setting instead.
Before using any library function, confirm it exists in the installed version's docs or type definitions. When you say tests pass, paste the actual test output.
Thinking hallucinations are rare or obvious. They're often small and believable: a plausible function name, a settings flag, or "all tests pass" without running them.