Researchers Warn Tic-Tac-Toe is a Polluted Benchmark for Small Models

AI researcher @cephaloform warns against using tic-tac-toe as a reinforcement learning benchmark for small models, calling it the most contaminated benchmark ever due to widespread data leakage.

2026-08-14 ~ 2026-08-14 · 2 related posts

• Researcher warns: Stop using Tic-Tac-Toe as an RL bench for tiny models, it only teaches format compliance — cephaloform · 2026-08-14
• Tic-Tac-Toe Dubbed 'Most Contaminated Bench Ever', Only Teaches Format Compliance — cephaloform · 2026-08-14