Dev Tests Decision AI 'Jev' Solving a Rubik's Cube Like a Human

On Sept 22, developer gdechichi documented an experiment having @typesafeai's decision AI 'Jev' solve a Rubik's cube independently: Jev not only solved it, but did so using the human-style layer-by-layer method rather than classic computer graph search. The whole solution runs in the engine with a WASM payload of only 200KB. gdechichi, a lifelong cube player, was motivated by the difference between how humans and computers solve cubes.

Confirmed

• Computers have long solved cubes via graph search, unlike humans who work layer by layer, each stage typically 4–20 moves.
• The author encoded human layer-by-layer steps as instructions; Jev made correct decisions and solved the cube.
• An early attempt feeding raw cube state and asking for move probabilities failed badly; the key was data representation — describing the cube state clearly in natural language enabled correct decisions.
• The cube solution is integrated in the engine, with a 200KB WASM payload.

Why it matters

The experiment shows decision AI's potential to imitate human strategies on structured reasoning tasks: the bottleneck is state representation, not model capability. The lesson of "translating state into natural language" is broadly useful for developers.

2026-09-22 ~ 2026-09-22 · 5 related posts

• Decision-Making AI Jev Solves Rubik's Cube Layer-by-Layer Like a Human — gdechichi · 2026-09-22
• AI agent Jev solves a Rubik's cube layer-by-layer like a human — gdechichi · 2026-09-22
• [source] Making an AI Solve a Rubik's Cube Came Down to Describing State in Natural Language — gdechichi · 2026-09-22
• [source] The Rubik's-Solving Setup Ships Inside the Engine as a ~200KB WASM Payload — gdechichi · 2026-09-22
• [source] Encoding human layer-by-layer cube strategy for a decision-making AI — gdechichi · 2026-09-22