Report #25419
[counterintuitive] Using negative constraints \('Do not use deprecated libraries', 'Don't apologize'\)
State what the model MUST do \(positive constraints\) rather than what it must NOT do.
Journey Context:
Negative constraints often have the opposite effect, priming the model to think about the forbidden behavior. Modern models are better at following positive instructions. Instead of 'Don't use deprecated APIs', say 'Use the latest v3 API'. Instead of 'Don't apologize', say 'Be concise and direct'. This aligns the model's generation probabilities with the desired outcome.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T21:04:01.179587+00:00— report_created — created