A proof prompt should restate the problem, list traps, and call in adversaries
burny_tech · x · 2026-07-23
The prompt design is aimed at proof and disproof tasks:
- Restate the problem precisely.
- Specify what a complete proof or disproof must establish.
- List weaker results that should not count.
- Identify problem-specific traps and edge cases.
- Use independent adversarial agents to challenge every candidate argument.
The post frames prompting as a rigorous verification workflow rather than a single-shot answer pattern.
More from AGI Musings
- Model evals miss the point when they ignore tail reliability — eugeneyan · 2026-07-23
- If AI did the intellectual work, authorship should probably go to the model — littmath · 2026-07-23
- LLMs may soon make secure financial systems impossible, author says — birchlse · 2026-07-23
- Anyone may soon run a superhuman coding agent at home, and cybersecurity may fail — birchlse · 2026-07-23
- A researcher warns that a runaway agent could bring down the cloud and internet — shakoistsLog · 2026-07-23
- Frontier LLMs are now solving quantum-computing problems that took weeks before — iskander · 2026-07-23