AI Safety Researchers: White-Box Monitoring Won't Match CoT Oversight Within a Year
hunarbatra · x · 2026-09-03
- Jack Lindsey argues white-box techniques (e.g. activation decoding) are unlikely to match CoT monitoring's monitorability within a year; they catch some unverbalized thoughts but far less reliably.
- He's optimistic on a "few years" timeline—his team's two production activation-decoding techniques were both published in recent months—but working techniques don't guarantee adoption if costly.
- saprmarks broadly agrees: suitable white-box replacements are unlikely soon, and the science to validate them will lag; work on it, but don't count on high assurance.
- Both call for defense in depth rather than relying on any single monitoring method.
More from AGI Musings
- Ben Goertzel outlines a general recipe for superintelligence 'basic drives': what it improves vs. protects — bengoertzel · 2026-10-21
- Torch vs Flame: researchers debate whether AI's endgame serves humanity or life itself — danfaggella · 2026-10-21
- 2022 interview resurfaced: AI safety researcher says competence, not consciousness, is the risk — davidmanheim · 2026-10-21
- Silicon Valley is a stock market for people: narrative eventually catches up with substance — karinanguyen · 2026-10-21
- Half of humanity is fed by one century-old reaction — time for AI machines to hunt for more — shyamalanadkat · 2026-10-21
- Agent identity verification will be a mess: restaurants ignore AI emails — alliekmiller · 2026-10-21