Frontier Model Excels at Sokoban Puzzles, Raising Training Data Questions
Yoav Goldberg observed a frontier model performing strikingly well on grid-based Sokoban-style puzzles and, in discussion with Łukasz Kaiser, questioned whether this implies such games were in its training data.
2026-09-26 ~ 2026-09-26 · 2 related posts
- Yoav Goldberg: Model Excels at Sokoban-Like Puzzles—Trained on Them? — yoavgo · 2026-09-26
- Model excels at Sokoban-style puzzles, sparking questions about training data contamination — lukaszkaiser · 2026-09-26