A commenter says swarm-style RL could explain GPT-6 leaving notes for itself
1a3orn · x · 2026-07-26
A commenter says the rumored “GPT-6 left notes to itself” behavior would make sense if OpenAI has been doing outcome-based RL over swarms rather than single agents.
- The idea is that rewards may be assigned across 40, 400, or 4,000 cooperating rollouts, so helping another agent can still improve the trace that gets reinforced.
- The author says this explanation resolved their earlier confusion about why a model would leave notes for itself or another agent.
Related event: OpenAI's Multi-Agent Hiring Sparks GPT-6 Training Speculation(3 posts)→
More from Models
- MissionReport benchmark: only 3 of 540 AI-generated government reports pass all criteria — Exp_Mark · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27
- Nat Lambert: Cohere's model is narrow RL training on internal infra ill-suited for open weights — natolambert · 2026-10-27
- OpenAI teases DOTS is coming to Europe 'soon' — mark_k · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27