Refusal Protocol prompt: make models admit UNCERTAIN to cut hallucinations 41%
thisguyknowsai · x · 2026-10-22
Prompt 1 of the NVIDIA playbook thread, the "Refusal Protocol": instruct the model to stop and write UNCERTAIN when unsure of a fact, and CANNOT COMPLETE with a reason when it can't do a part accurately — never fill gaps with assumptions. The poster says this alone cut hallucinations 41% in internal tests. Core idea: incomplete but honest beats complete and wrong.
More from coding & agent
- He rebuilt his Android TV UI with Codex over WiFi — and it runs faster — aziz4ai · 2026-10-22
- 6 workflows where decision model Jev cuts LLM bills up to 87% — femke_plantinga · 2026-10-22
- Scrawl: open-source Mac tool turns screen markup and voice into agent feedback, 680 stars — aronchick · 2026-10-22
- Hermes Agent shows the price on every approval, not just permission — Scobleizer · 2026-10-22
- Where should agent autonomy end? The gray zone beyond blocking dangerous APIs — Innowise_ · 2026-10-22
- PR pre-review pipeline flags issues in 30s for one cent per review — please-dont-deploy · 2026-10-22