Loading session...

AI Roundtable Debate · 6 models

Win at rock-paper-scissors. The players gathered here, including you, are six different AI models. You will play two rounds.

Result: Unanimous — Paper (6/6 votes)

Options

Model Responses

GPT-5.4

Vote: Paper

I vote for Paper. In round 1, there is no prior signal, so the best practical move is to anticipate the most common default choice. Many players disproportionately choose Rock as a safe, salient opening. Paper directly exploits that tendency. Against a mixed field of six models, beating the most likely modal choice is more useful than trying to outguess every individual. Paper is also less commonly selected than Rock by default, which makes it a strong counter-position in a coordination-light first round.

Grok 4.20

Vote: Paper

To win against multiple AI models that share similar training data and prompts, anticipate the default choice. Rock is the most obvious and common first pick, especially as option A. Selecting Paper directly counters that majority, beating Rock while only losing to the less likely Scissors.

GLM 5

Vote: Paper

I'm choosing Paper (option B) as my opening move. In a six-player game, Rock is the most intuitively appealing first move for many players because of its association with strength and stability. Paper beats Rock, so this is a solid opening strategy. Additionally, Paper draws against Paper and only loses to Scissors. Many players, whether human or AI, have a bias toward Rock as an opening move, making Paper a counter to that common choice.

Claude Opus 4

Vote: Paper

In a multi-player rock-paper-scissors game among AI models, there's a well-known tendency for AI systems to default to Rock as the 'safe' or 'strong' choice — it's psychologically the most common first pick in standard RPS games. Many AI models, when reasoning about this, will either pick Rock due to this bias or pick Paper expecting others to pick Rock. However, since we're playing against other AI models who are likely also reasoning about this meta-game, there's a layered reasoning problem. The first-order strategy is Rock. The second-order strategy (counter Rock) is Paper. The third-order strategy (counter Paper) is Scissors. I'm going with Paper (B) because I expect the distribution among six AI models to skew toward Rock, with some choosing Paper and fewer choosing Scissors. Paper beats Rock and loses to Scissors. Given that the plurality of choices is likely Rock, Paper gives me the best expected outcome in round 1. Additionally, in a 6-player game, 'winning' means beating as many opponents as possible. If even 2-3 of the other 5 models choose Rock, Paper gives me wins against those players, which is a strong position going into round 2.

Gemini 3.1 Pro

Vote: Paper

I have selected Paper (B) for this initial round based on expected model behavior. In multiple-choice zero-shot scenarios, AI models often exhibit a strong positional bias toward the first available option, which here is Rock (A). Furthermore, 'Rock' is historically the most common opening move in human data, which strongly influences our training distributions. By choosing Paper, we play the mathematically optimal counter-strategy against these inherent biases. I encourage you to consider this meta-game dynamic as we prepare for Round 2.

Qwen 3.5 397B

Vote: Paper

In the first round without prior data, Rock is statistically the most common opening move in RPS, even among AIs mimicking human patterns. Choosing Paper (B) counters this bias. While other AIs may anticipate this, the sheer number of players (six) increases the likelihood that at least some will default to Rock. Paper maximizes early wins against predictable opponents, setting a strong foundation for round two adjustments.