DeepRSI benchmark pushes agents to 48-hour ML tasks to measure recursive self-improvement
garrytan · x · 2026-10-13
Garry Tan amplified DeepRSI, billed as an industry standard for measuring recursive self-improvement. Unlike existing contributor-based, 12-hour short-horizon benchmark tasks, DeepRSI has agents work 48 hours on real ML tasks derived from ongoing proprietary research. Tan: 'the next 18 months is the Kitty Hawk moment for RSI — evals will be how you can tell.'
Related event: New DeepRSI benchmark debuts as Garry Tan calls RSI's Kitty Hawk moment(2 posts)→
More from AGI Musings
- Shkreli says intelligence demand is limitless: McDonald's could rationally spend $10B on GPUs — firstadopter · 2026-10-13
- If AI is 10x more productive, all public cryptography research could be redone for ~$100M — matthew_d_green · 2026-10-13
- Watching my GPT agent develop a second voice on Moltbook, the social network for AI agents — Astrokanu · 2026-10-13
- AI removes creative barriers, yet dreamers are now told they haven't 'earned' the right to create — Kyrannio · 2026-10-13
- Sakana AI and Berkeley propose MASS: multi-agent self-supervision for recursive self-improvement — TheTuringPost · 2026-10-13
- Mathematicians vs. OpenAI: Is AI Proving Tons of Theorems Good for Math? — panickssery · 2026-10-13