Match recurring AI output failures to appropriate prompt patch patterns.
Generic answer: context patch or example patch? Use context patches when the output is plausible but not situated. Unsupported factual claim The answer includes a number, policy, date, or citation that may not be in the source. Patch pattern Source-only rule + quote-first extraction + claim citation + Unknown fallback. Ask the model to double-check facts without defining evidence. It changes the output contract from completion to supported completion. Format drift across repeated runs The answer is readable but changes fields, labels, or sections. Exact schema + allowed values + missing-data behavior + no extra keys + one valid example.…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in