FrontierThe story, in brief
Dota 2 with large scale deep reinforcement learning
OpenAI's AI beat the world's best Dota 2 players. Here's what that means for reasoning.
The KeyNews take
Why it matters
OpenAI demonstrated large-scale deep reinforcement learning can master complex multi-agent strategic games, validating RL as a path to more capable AI systems. This became a foundational case study for scaling training approaches.
The key facts
5 to knowOpenAI Five defeated top professional Dota 2 teams in 2019
Large-scale deep reinforcement learning applied to complex real-time strategy game
Multi-agent coordination and long-horizon planning capabilities demonstrated
Published December 2019 — landmark AI capability milestone
Reinforcement learning scaled to game environment with >20,000 possible actions per turn
Go to the source
OpenAI Blogopenai.com