Definition
Plain language
The famous AI that beat top human professionals at heads-up poker, built with millions of hours of self-play computation.
As stated in the literature
A 2017 superhuman heads-up no-limit Hold'em system from CMU built on counterfactual regret minimization, requiring over fifteen million CPU core-hours; the canonical solver-based poker AI that PokerSkill positions itself against.
Why it matters: It set the bar for superhuman poker while showing how enormous the computing cost was, which is the standard newer, lighter approaches are measured against.
For example, this system beat top human poker pros, but only after running the equivalent of millions of hours of computer time to prepare.
Heard on the show
“The famous superhuman bots — Libratus is the headline one — were built on an algorithm called counterfactual regret minimization.”Episode 100 — How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert