The 8 most valuable data points AI labs should share to measure recursive self-improvement
nitarshan · x · 2026-09-24
In a thread, Cheryl (Wooo) proposes how to measure labs' RSI (recursive self-improvement):
- Core claim: RSI would likely accelerate AI capability growth, so externally verifiable metrics are needed
- She lists the 8 most valuable data points labs should share; the first: report performance on diverse benchmarks for the latest internally deployed models
A substantive AI-governance proposal: an actionable disclosure checklist for frontier lab transparency.
More from Safety
- "We used AI to polish sentences" is a bad disclosure, economist argues — paulnovosad · 2026-09-24
- A 300-word 100%-AI threshold: economist proposes a testable AI-content rule — paulnovosad · 2026-09-24
- Redwood Research: latent reasoning architectures would undermine CoT oversight — jammastergirish · 2026-09-24
- Economist proposes journals use Pangram to detect AI-written papers, with 2-year submission bans — paulnovosad · 2026-09-24
- Using the Jev model for offensive security: OPSC ranking and sensitive file detection in Mythic — dyn___ · 2026-09-24
- New research finds open-weight LLMs can experience pain, gaslighting hurts more than insults — _ArkAngel_ · 2026-09-24