AI pokerbot hackathon: winner distills its hints into weights as bots lie and prompt-inject each other

rhythmrg · x · 2026-11-21

A pokerbot hackathon by appliedcompute and ValsAI gave every team the same model and 24 hours; bots could talk, lie, and prompt-inject each other at the table. One team found a deterministic Python script outplayed the LLM and trained the model to copy it; another taught theirs to stay completely silent. The winner distilled its hints into the weights. A fun showcase of agent game-playing and adversarial behavior.

Original post →

More from coding & agent

coding & agent channel →