Open questions in action-only AI monitoring: intelligence gap, test-time compute, sync limits
xeophon · x · 2026-09-03
In a discussion on action-only monitors (watching model actions rather than chain-of-thought), xeophon notes that the open-source verifiers framework already supports synchronous action-only monitoring.
The quoted thread lists key open research questions:
- How large can the intelligence gap between monitor and main model become before it turns critical?
- How much test-time compute should the monitor spend to catch misalignment?
- Can truly synchronous action-only monitoring be implemented, and if not, what temporal gap is allowable?
More from Safety
- Center for AI Safety Publishes Full Syllabus for Intro to ML Safety Course — ArtificialOther · 2026-10-21
- Ben Goertzel outlines a general recipe for superintelligence 'basic drives': what it improves vs. protects — bengoertzel · 2026-10-21
- America.gov launches AI portal unifying 29,000 government sites with fraud reporting — soleio · 2026-10-21
- EvilWAF: open-source transport-layer MITM proxy that bypasses modern WAF behavioral detection — godfrey_rono · 2026-10-21
- 2022 interview resurfaced: AI safety researcher says competence, not consciousness, is the risk — davidmanheim · 2026-10-21
- Domingos: GDPR purpose-limitation makes LLM pretraining illegal in Europe — pmddomingos · 2026-10-21