Gemini 4 safety report shows AI hits 80% reliability on only ~15-minute AI R&D tasks
tobyordoxford · x · 2026-10-22
Toby Ord highlights that Google DeepMind's Gemini 4 Safety Report (p.78) published an internal analysis of AI for AI R&D: the model's 80% success horizon on real AI R&D tasks is only about 15 minutes. He reads this as a genuine update confirming how surprisingly short the tasks AIs can do reliably and autonomously remain — notably the first frontier-lab release validating METR-style horizon curves with internal AI R&D usage data.
More from AGI Musings
- Tyler Cowen: Reading AI Writing Is an Option on Future Connections Worth Near Zero — paulnovosad · 2026-10-22
- ChatGPT, Claude and Grok all answer p(doom) with ~10%, an odd cross-model consensus — maksym_andr · 2026-10-22
- Will math papers admit "the author cannot verify these details" in acknowledgements? — pickover · 2026-10-22
- What MLB's ABS challenge system and NBA's sensor ball teach us about handing decisions to machines — lmoroney · 2026-10-22
- Dario Amodei's stark take: AI threatens identity and the tax base, not just wages — coherence · 2026-10-22
- Wiblin's plea to AI lab staff: reassess now whether your work delays or accelerates RSI — geoffreyirving · 2026-10-22