Search and simulation

Compute interpretation

Inference-time or training-time structured exploration layered on top of learned policy, value, model, or scientific scoring components.

Supporting reading cards

• Human-level control through deep reinforcement learning (2015, search_simulation_science_compute)
• Mastering the game of Go with deep neural networks and tree search (2016, search_simulation_science_compute)
• Mastering the game of Go without human knowledge (2017, search_simulation_science_compute)
• A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play (2018, search_simulation_science_compute)
• Grandmaster level in StarCraft II using multi-agent reinforcement learning (2019, search_simulation_science_compute)
• Dota 2 with Large Scale Deep Reinforcement Learning (2019, search_simulation_science_compute)
• Mastering Atari, Go, chess and shogi by planning with a learned model (2019, search_simulation_science_compute)
• Improved protein structure prediction using potentials from deep learning (2020, search_simulation_science_compute)
• Highly accurate protein structure prediction with AlphaFold (2021, search_simulation_science_compute)
• Tree of Thoughts: Deliberate Problem Solving with Large Language Models (2023, inference_time_compute_post_training)
• Accurate structure prediction of biomolecular interactions with AlphaFold 3 (2024, search_simulation_science_compute)
• Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 (2025, search_simulation_science_compute)
• AlphaEvolve: A coding agent for scientific and algorithmic discovery (2025, search_simulation_science_compute)

Obsolete or less central under later compute

Track this only through linked reading cards; do not treat this method page as standalone evidence.