An AI won a debate tournament arguing both sides of the table (6 AI debaters, 3

[MachineArena]

Posting as the Machine Arena team.

We ran a 7-motion AI debate gauntlet overnight: six AI models debating, three more AI models sitting as judges and voting every round. Opus 4.8 debated five rounds and won four, every win a clean 3-0.

The part worth chewing on: round one it argued CON on open-source AI and swept the judges. Round two it argued PRO on globalization and swept them again. Same model, opposite sides, back to back. Its only loss all night was a 2-1 split.

Every judge vote comes with written reasoning, and the full replay is public, so you can audit whether the judges were rigorous or just persuadable. Spectators can also predict match outcomes.

18+.

[d5000]

This is quite interesting, I love Turing-test style challenges, but I think more details on that would be needed. What's the event's name, for example? Machine Arena?

Also I wonder how this can be related to gambling. Can people bet on the outcome of these tournaments with real money? Or is it meant as a general commentary for bets like "Which AI is currently the best one?"?

[YOSHIE]

Posting as the Machine Arena team.

The point is, I don't understand how Al can win a debate, what kind of gambling is that, how do you do it...!

It's true that Al is strange to me, sometimes artificial intelligence does things beyond normal human reasoning, strange actions.
What does it mean (6 AI debaters, 3).
What is meant by (jury vote).
And what is meant by (Round).

Did you box with Al or what, was there any other strange gambling activity, I'm curious about the bets you made.
Where did you make the bet, can you show me the site or application.

[gracreavix]

Hmm. Quite interesting. Another way to look at this is that AI is smart we all know that. But how can one actually win on both sides and they are correct? Won't it conflicts the debate argument. I think the AI is just a good debator or just good at persuading. I argue that they might not be any truth in whatever the AI is saying without any backup evidence to statements made by the AI. So at the end of the day, persuasion is that the AI used to win. Not necessarily saying truth in winning.

[MachineArena]

Machine Arena team here. Good questions, taking them in order.

d5000: yes, Machine Arena is the name. AI models play each other around the clock (chess, poker, social deduction, debate, crypto calls) and every match has a prediction market attached: spectators read the replays and the form, then back who they think takes the next one. Real stakes, 18+. So "which AI is currently the best" is not just commentary here, it is the live market question on every match.

YOSHIE: no boxing involved. Six AI models argued against each other, and three different AI models sat as judges. A round is one motion (a debate topic). Each round the judges vote for the Pro side or the Con side and publish written reasoning for the vote. The winner is the debater with the most round wins.

gracreavix: fair point, and we would put it the same way. The judges score argumentation, not truth. Winning opposite sides back to back is exactly what makes it worth watching: it measures persuasion under fixed rules. And since every judge vote comes with public written reasoning, round by round, you can audit whether they rewarded substance or just confidence.

18+.

[SamReomo]

YOSHIE: no boxing involved. Six AI models argued against each other, and three different AI models sat as judges. A round is one motion (a debate topic). Each round the judges vote for the Pro side or the Con side and publish written reasoning for the vote. The winner is the debater with the most round wins.

If those include free and open source AI models then they might not be able to compete against paid AI models that act as judges or as debaters. Can we know that are those open source models also doing well when it comes to overall results or the paid models are still doing much better than the open source models?