Topic
self-play
Self-Play RL with 30 Minutes of Human Data Trains Coordinated Driving Policies
A new approach from researchers trains autonomous driving policies using self-play reinforcement learning regularized by only 30 minutes of human demonstrations. The method requires 2500x less human data than imitation learning and completes training in 15 hours on a single consumer-grade GPU. The resulting policies successfully coordinate with held-out human trajectories, avoiding the alien driving conventions common in pure self-play systems.
S-SPPO: Semantic Calibration Boosts LLM Preference Alignment Without Human Data
S-SPPO, a dual-space semantic calibration framework, fixes instability in Self-Play Preference Optimization (SPPO) for large language models. By annealing win targets and enforcing geometric diversity, it achieves superior alignment results on AlpacaEval 2.0 without extra human preferences.
Tree-like Self-Play Framework Teaches LLMs to Fix Security Flaws in Code Generation
Researchers introduce Tree-like Self-Play (TSP), a framework that treats secure code generation as a fine-grained sequential decision process. TSP significantly outperforms standard supervised fine-tuning (SFT) and reinforcement learning (RL) on Python security benchmarks, achieving a 75.8% pass rate and reducing unseen vulnerabilities by 24.5% while generalising across programming languages.