Leaked OpenAI loss ladder: unknown capabilities await at 0.68-0.7
tallmetommy · x · 2026-11-23
OpenAI reportedly published its internal loss ladder mapping loss levels to capabilities: GPT-3.5 at 0.99 could chat, GPT-4 at 0.91 could use chain-of-thought, and math breakthroughs appeared around 0.76. The next window is 0.68-0.7, and OpenAI's chief scientist says he doesn't know what will show up there.
More from Models
- Thinking Machines releases Inkling 2 Small, scoring 31 on AA Intelligence Index — _Hao_Zhu · 2026-11-23
- Developers heap praise on Gemini 4 Neon's 'most tasteful' writing — gajesh · 2026-11-23
- METR-fit extrapolation implies 66-hour task horizons for 6.2 Astra, 128 hours for internal model — eliebakouch · 2026-11-23
- $200 Google AI Ultra subscription quietly includes a 96GB Blackwell GPU for 10 days — kieranklaassen · 2026-11-23
- Tesla FSD insider: end-to-end is settled, winning hinges on data diet quality — Singularitarian · 2026-11-23
- Google sweeps top 3 on Arena text leaderboard as Gemini 4 Carbon scores 1527 — demishassabis · 2026-11-23