Real-world time horizons at OpenAI/GDM far shorter than models predict, says Ramez
ramez · x · 2026-11-12
Replying to Ben Todd, ramez pushes back: actual time horizons observed inside OpenAI and Google DeepMind are dramatically shorter than claimed, and at OpenAI they don't appear to be rising exponentially. Empirical evidence diverges from the model-based predictions — a direct challenge to METR-style doubling narratives.
Related event: Debate Erupts Over Credibility of AI Progress Forecasting Models(18 posts)→
More from AGI Musings
- Future generations will find our debates over whether AI should make art hilarious — VraserX · 2026-11-12
- He tried to prove Crunchbase misses deals — it covered 41 of 45, logged 1 day early — YvesMulkers · 2026-11-12
- Critics slam Amodei's cure-all prediction: AI bubble draining investment from real biotech — AlexTensor · 2026-11-12
- AI can write all the code, but the hard part of software engineering was never the code — bendee983 · 2026-11-12
- ChatGPT co-creator: LLMs are built for assistance, not automation — SimplyAnnisa · 2026-11-12
- Meta's François Fleuret sketches an agent-native Rust OS whose GUI is filtered by an LLM — francoisfleuret · 2026-11-12