ChatGPT nails a backgammon dice probability problem at 47% odds

ivan_bezdomny · x · 2026-09-03

The author used ChatGPT to compute backgammon dice odds: at roughly 47% win probability (near a coin flip), the needed rolls were 22, 33, or 34+ — and the model got it right.

He also shared his aggressive match style (short matches, two gammons in five) and noted it's a small marker of progress in probabilistic reasoning, wondering how long ago models would have failed this problem.

Original post →

More from Models

Models channel →