https://styleofwong.cn/en/tags/alphazero/Recent content in AlphaZero on Wang Huijiu's Sitehttps://styleofwong.cn/apple-touch-icon.pnghttps://styleofwong.cn/apple-touch-icon.pngHugoen-USSun, 28 Jun 2026 19:30:00 +0800https://styleofwong.cn/en/posts/balatro-rl-design/Sun, 28 Jun 2026 19:30:00 +0800https://styleofwong.cn/en/posts/balatro-rl-design/If we want to achieve a 99% Ante 8 clear rate on Balatro White Stake, how should the RL algorithm be designed? This article presents a concrete technical blueprint from game-mechanics modeling, state/action spaces, reward shaping to algorithm selection, and explains why pure PPO won’t work and why a hybrid MCTS+RL architecture is necessary.https://styleofwong.cn/en/posts/rl-breakthroughs-alphazero-dota/Sun, 28 Jun 2026 19:00:00 +0800https://styleofwong.cn/en/posts/rl-breakthroughs-alphazero-dota/AlphaZero and OpenAI Five are two milestones in reinforcement learning. What walls did each of them hit, and what engineering and algorithmic innovations knocked those walls down? This article dissects them with a pyramid structure: three shared bottlenecks — data, stability, and scale — and how each project broke through them.