OpenAI wants labs to report recursive self-improvement metrics, but its own definitions drift
tallmetommy · x · 2026-11-19
OpenAI proposes that labs be required to report recursive self-improvement indicators. The poster agrees in principle but notes problems in OpenAI's own reporting: a harness tweak depressed token spend for weeks, and the code metric was redefined since September.
- Takeaway: mandatory metrics need frozen definitions and an outside auditor to be meaningful
More from Safety
- Analysts: frontier intelligence vs enterprise readiness gap is widening, big security upgrade cycle coming — BenBajarin · 2026-11-19
- Google report details autonomous AI hacking system, but no confirmed breaches — TruthIsAllYouNeed_ · 2026-11-19
- Trump's China data center curb advances despite Amazon, Google, Meta lobbying — pstAsiatech · 2026-11-19
- Nadella: Much Critical Infrastructure Is 'Unpatchable' Amid AI Cyberattack Fears — TechCrunch AI · 2026-11-19
- WireWatch measures security of proprietary network encryption across the Android ecosystem — jedisct1 · 2026-11-19
- Google CISO Phil Venables: We're in the Biggest "Uncanny Valley of Security" Yet — philvenables · 2026-11-19