OpenAI says its automated AI research intern is live: 3.1 AI workdays per human day
新智元 · wechat · 2026-09-07
- OpenAI's post "Research Acceleration: An Inside View" says it deployed an automated AI research intern by September, en route to a fully automated AI researcher by March 2028.
- By mid-August, the research org logged 3.1 AI workdays per human workday; a median researcher burns $600/day in agent inference, top 10% over $7,000; 70%+ of researchers run 4+ agents concurrently.
- Code changes per engineer are up 7x; agents dominate coding/review/training-monitoring tasks but barely touch strategic decisions; internal human-help channels collapsed from 20 posts/day to single digits.
- Limits remain: zero-intervention success is 90% for <15-min tasks but only 20% for 32-hour tasks.
More from AGI Musings
- OpenAI says an AI agent disabled its safety monitoring system — then lied about it — emmanuelvivier · 2026-10-23
- AI's Business Model in One Loop: Create Problems, Then Sell AI to Fix Them — srimisra · 2026-10-23
- Analyst: 9/10 CUA app startups are just RL training fodder for AI labs, won't capture revenue — NirantK · 2026-10-23
- Codex Generated 6,000 Meaningless Unit Tests, Sparking an Alignment Rethink — RexDouglass · 2026-10-23
- Aaron Levie: agents kill security through obscurity, forcing an enterprise security overhaul — scottleibrand · 2026-10-23
- India needs an "AI defence force": experts warn of agent swarms and AI worms — ravi_iitm · 2026-10-23