Why do Astra and Sol 6.1 perform so well? Multi-agent RL and epistemic vigilance
scaling01 · x · 2026-11-06
scaling01 speculates on why Astra and Sol 6.1 perform so well: general reasoning gains from 'dirty loopers', or new multi-agent RL—specifically safety training for cogsec, termed 'epistemic vigilance'. He also suspects OpenAI has RL environments for all such games and analogous message-board/web scenarios.
More from Models
- John Schulman's Inkling 2 Small: capabilities-focused RL trains misalignment by default — SonglinYang4 · 2026-11-06
- Baidu open-sources ERNIE 5.5 Flash Base, a 280B-parameter MoE foundation model shipped without post-training — Baidu_Inc · 2026-11-06
- All 22 DeepSeek V4.1 Flash endpoints statistically in line, most consistent model in the Index — ArtificialAnlys · 2026-11-06
- All 22 DeepSeek V4.1 Flash endpoints match reference — 3.9-point range, most consistent model in the Index — ArtificialAnlys · 2026-11-06
- GLM-5.3 has the most endpoints in the Accuracy Index but widest spread: 67.2 to 99.3 across 26 endpoints — ArtificialAnlys · 2026-11-06
- Artificial Analysis launches Endpoint Accuracy Index v1.1: nearly 1 in 4 serverless endpoints score below reference — ArtificialAnlys · 2026-11-06