Hallucination

A hallucination is when an AI model states something false with complete confidence: an API that does not exist, a function signature it invented, a fact it made up. Models are built to produce plausible-sounding output, and plausible is not the same as correct, so they will fill a gap with confident fiction rather than admit they do not know.

For a founder building with AI this is the failure mode to design around, because the model never signals which claims are solid and which are guessed. Say you're building a feature with Claude as your coding agent: it can call a library method that was never real, or claim a test passed when it never ran it. Or say you ask Perplexity to research a competitor and it produces a clean stat with a citation that, when you open it, says nothing of the sort.

The defence is never to trust output on its face, verify it against the real thing: check the actual API docs, run the code, read the row in Supabase yourself, drive the feature on the live surface. Good context cuts hallucination because the model has the real facts in front of it instead of guessing, which is exactly why feeding it the right files matters. Treat every confident claim from a model as a lead to verify, not a fact to bank.