Ramez: OpenAI's Productivity Gain Per ECI Point Appears to Decline Over Time
ramez · x · 2026-11-09
Analyst Ramez shared an unpublished finding: in OpenAI's monthly research-acceleration data, the compounded productivity gain per estimated ECI point appears to drop over time. Charted rolling three-month windows show gains sliding from 8.2% to 5.7%, 1.7%, then 1.4% (coding-agent tokens only, ECI gain 17.53/year, R²≈0.83–0.88 across fits). He caveats the smooth-ECI-growth assumption and calls for longer, more precise empirical data from all labs; in the thread he concedes related criticisms are fair and says he's refining the model.
More from AGI Musings
- Spotting AI slop on X: check quote-tweet comment sections from ~10K-follower accounts — MaziyarPanahi · 2026-11-09
- AI's New Caste System: Your Rank Is Decided by Compute, Not Birth — FinanceYF5 · 2026-11-09
- Models can be trained to talk to each other in inscrutable neuralese text — Sauers_ · 2026-11-09
- a16z essay argues AI's next economy shifts from inference to execution, a bigger market — AccBalanced · 2026-11-09
- "AI Forces You to Confront the Replaceability You Avoided All Your Life" Goes Viral — Yamapama · 2026-11-09
- Learning Abstractions: Why Immediate Concept Acquisition Is AI's Next Frontier — jmugan · 2026-11-09