Short answers, pulled from the story.
Pluribus is a computer poker player that became the first AI to beat professional humans in a complex multiplayer competition. Built by Facebook's AI Lab and Carnegie Mellon University and published in 2019, it plays no-limit Texas hold'em against multiple opponents simultaneously.
Pluribus was built by researchers at Facebook's AI Lab and Carnegie Mellon University. They published their results in 2019.
Training Pluribus' base strategy took eight days and cost about $144 at market computing rates. The developers noted this was much smaller than what contemporary superhuman game-playing milestones, such as AlphaZero, had required.
Pluribus avoids "limping," the practice of calling the big blind rather than raising or folding. It also uses "donk betting," ending one round with a call and then opening the next by betting, more frequently than professional players typically do.
Playing five professionals simultaneously in no-limit hold'em, Pluribus won an average of $5 per hand and approximately $1,000 per hour. Across all competitions it averaged over 30 milli big blinds per game, which Facebook described as a "decisive margin of victory."
The developers declined to release Pluribus' source code because they feared it would be used to surreptitiously cheat against human players in online poker matches. A bot this effective running undetected posed a specific risk they were not willing to create.