Self-play RL on one desktop beats Warcraft III's insane built-in AI
ToraxXx · reddit · 2026-10-19
warlock.ai published a WarcraftSim project writeup: the author trained an agent with self-play reinforcement learning (policy v267) on a single desktop to beat Warcraft III's built-in "Insane" AI.
The accompanying dashboard shows a live match — the learner (red Nightelf) vs the built-in insane AI (blue) — with reward curves, win estimates, resource (gold/lumber) comparison, and per-unit policy entropy, as the learner gradually overtakes the AI economically and militarily.
The simulation environment and training pipeline are open, making this a reproducible case of applying modern RL to a complex legacy RTS, with implementation details in the blog.
More from Research
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23
- Perceptic launches PercepticOS, an AI workspace for 'scientific superintelligence' — nathanbenaich · 2026-10-23
- EPFL's Bosselut to talk long context as test-time learning at ELLIS seminar — andre_t_martins · 2026-10-23