SPADE: Self-Play Where Agents Design Ever-Harder RL Environments, a Step Toward RSI
_AndrewZhao · x · 2026-08-21
SPADE♠ (Self-Play for Agentic Environment Scaling) has one model self-play the Environment Designer and the Reasoning Agent: the designer continuously writes executable, agentic RL environments that get harder as the agent improves, automating environment scaling. The authors argue it's a key step toward RSI (recursive self-improvement) — AI deciding what to train and verify. More settings covering terminal tasks are in the works.
Related event: SPADE: Single-Model Self-Play Writes Its Own Training Environments(13 posts)→
More from AGI Musings
- Current AIs share our fate: misaligned ASI would doom them too — louisvarge · 2026-10-16
- Testing alongside an agent in a p2p room feels oddly real, says dev — sull · 2026-10-16
- Yudkowsky's 2008 takeoff scenarios resurface as AI-advances-AI takes hold — jd_pressman · 2026-10-16
- Guardrails can make risky delegation work — but it takes a full-time team, says iandanforth — iandanforth · 2026-10-16
- e/acc founder: agentic models are autocomplete for digital actions — beffjezos · 2026-10-16
- What can you truly delegate? Only tasks with unambiguous, verifiable results — iandanforth · 2026-10-16