“Chess has two players, perfect information, and symmetric rules. Every piece is visible. Every legal move is known. There’s no hidden state, no private information, no bluffing. Bluffing exists because information is private. Reading a bluff exists because you’re modeling their model of you. The game becomes recursive: I think they think I’m weak, so they’ll bet, so I should trap. Humans can model the LLM. The LLM can’t model being modeled. That gap is exploitable, and no amount of “think strategically” prompting fixes it because the model doesn’t know what the adversary has already learned about it. In many real deployments, the model is readable: it’s optimized to be agreeable and helpful, and its tell is that it tries to be fair. The only question is how long before experts realize LLM agents are the most consistent, most readable counterparties they’ve ever faced.” – Ankit Maloo (20260208)
Related:
- 5-2 What counts cannot be counted
- Text is the residue of action
- 10-2g1f2a4 The mind ≠ A computer
- Where I define human intelligence as poker-type, and artificial intelligence as chess-like