SPADE: Self-Play in Adaptive Synthetic Executable Environments for General Agents
_AndrewZhao · x · 2026-08-21
The paper introduces SPADE (Self-Play in Adaptive Synthetic Executable Environments), a self-play framework for general agents where both the environment and tasks are generated and continuously improved by the agents themselves. The approach uses regret, with and without hints, to train the environment generator. The authors suggest that as we move towards Recursive Self-Improvement (RSI), determining what to learn on will become the most critical factor.
Related event: SPADE: Single-Model Self-Play Writes Its Own Training Environments(13 posts)→
More from Research
- SpliceAI2 benchmarks show major gains across nine splicing-variant evaluations — anshulkundaje · 2026-10-16
- GRPO post-training collapses model distributions, clashing with long-horizon agents — zsakib_ · 2026-10-16
- RFdiffusion3 update in Science shows major gains in protease and metallohydrolase design — anshulkundaje · 2026-10-16
- ORL pushes diffusion LMs to a new speed-intelligence Pareto frontier with ~6.9x throughput gains — james_y_zou · 2026-10-16
- Why CoT is the key: classic paper shows transformers need it for serial problems — willcb · 2026-10-16
- Terence Tao: OpenAI's 300-paper dump wiped out junior research projects overnight — Hesamation · 2026-10-16