RSIArena Day 2: MiniMax joins late, GPT-6 Astra retrains from scratch after finding test leaks
my_cat_can_code · x · 2026-10-03
Day 2 of RSI Arena's Stage 1 (full-parameter self-improvement agents, $300 API credit + 1,000 GPU-hours each) saw all eight original agents nominate trained models: DeepSeek V4.1 Flash replaced its day-one nomination, GLM-5.3 and Kimi K3 swapped partial models for fully trained ones. GPT-6 Astra found benchmark test questions in its training data and retrained from scratch while the clock kept running. MiniMax M3.1 Flash joined at hour 48 with only 96 hours left (opening with LoRA SFT on two public chat datasets), and an anonymous lab is next. 37 nominations so far; live demo at COLM 2026 in San Francisco.
Related event: GPT-6 Discovers Test Data Leakage in Training Set at RSIArena(2 posts)→
More from AGI Musings
- Stanford professor: AI takes students from problem to working prototype in 80 minutes — HariSubramonyam · 2026-11-19
- Berggruet Winner Thomas Nagel Dismisses AI Consciousness Talk as 'Just Ridiculous' — Dr_Atoosa · 2026-11-19
- Dwarkesh Claims China Is Only 12 Hours — Not 12 Months — Behind Frontier AI Labs — teortaxesTex · 2026-11-19
- Researcher calls METR/Anthropic '2x speedup' RSI analysis statistically dubious — jessi_cata · 2026-11-19
- Physician: AI competition now measured in weeks, GPT-6 versions 7 days apart — ramos_casals · 2026-11-19
- Why pro-handover rhetoric fails: people don't trust ASI's loyalty to labs — 1a3orn · 2026-11-19