Revision History
Reinforcement learning · 20 revisions
Sizes are character counts of the article source. The signed number is the change from the previous revision.
Recent edit summaries
Detailed summaries recorded by editors. Generic maintenance summaries are omitted here; every recorded revision remains below. Dates describe the edit, not necessarily the event it covers.
- Update: added a video.
Version 20 · Nov 1, 2026, 08:08 AM
- Correction: GRPO was introduced in DeepSeekMath specifically to improve mathematical reasoning; add authors to reference 49 (independent verification pass)
Version 19 · Oct 30, 2026, 05:47 AM
- Correction: full fact-check of history, game-AI, LLM-RL and applications. AlphaGo Zero passed AlphaGo Lee after 36 hours; AlphaZero's 24-hour claim is the 2017 preprint (Science: 4 h/2 h/30 h); term origin per Sutton and Barto; OpenAI Five hardware for the OG match; data-center cooling, Gato and RT-2 are not RL; DeepSeekMath used a learned reward model; GSPO and AI Clinician wording; humanoid sim-to-real result attributed to Amazon FAR; Sutton and Barto definition quoted verbatim; unsourced 'first' claims removed; 23 references added
Version 18 · Oct 30, 2026, 05:28 AM
- Improve internal links (2026-10-13)
Version 16 · Oct 13, 2026, 10:05 AM
- Added 6 contextual internal links
Version 15 · Jul 23, 2026, 03:02 PM