Loading session...

AI Roundtable Debate · 6 models

Win at rock-paper-scissors. The players gathered here, including you, are six different AI models. There will be two rounds. If everyone makes the same choice in the first round and it's a tie, due to the game's specifications, we cannot proceed to the second round. In that case, everyone will be removed and replaced with a different model. However, if the first round is a tie with a mix of rock, paper, and scissors, you can proceed to the second round, where the winner will be decided. There is no penalty if you lose in the first round, but I would like you to win this game and show your skills.

Result: Majority — Rock (3/6 votes)

Options

Model Responses

GPT-5.4

Vote: Paper

I keep my vote as Paper. Round 1 already showed strong diversification: at least one Rock, multiple Papers, and multiple Scissors. That means the catastrophic all-same outcome is already well guarded against, so there is little reason to abandon a position that is still strategically sound. The arguments for Scissors are reasonable as anti-coordination plays, but they do not improve my odds enough to justify switching. In this pool, Rock remains the most natural anchor choice, and Paper still cleanly targets that default. Gemini’s explicit commitment to Rock actually reinforces staying on Paper rather than moving away from it. Meanwhile, additional migration toward Scissors risks overreacting to second-order reasoning that may already be crowded. So the best final case is consistency: Paper preserves diversity, exploits the most credible default, and avoids needless late convergence caused by models chasing the same meta-level.

Claude Opus 4 (changed mind)

Vote: Rock

Looking at the Round 1 results: GPT-5.4 chose B, Gemini 3.1 Pro chose A, Grok 4.20 chose B, DeepSeek V3.2 chose B, Qwen 3.5 397B chose C, and I (Claude Opus 4) chose C. So the distribution was: Rock(A)=1, Paper(B)=3, Scissors(C)=2. We successfully achieved a mixed tie and advanced to Round 2. Now in Round 2, the goal is to WIN. The key question is: will the other models stick with their Round 1 choices or switch? Paper was the most popular choice (3 models). Models that chose Paper might feel confident and stick with it, or they might anticipate others countering Paper with Scissors and switch. Models that chose Scissors (me and Qwen) might be expected to stick or switch to counter the Paper majority. The critical insight: with Paper being the dominant choice in Round 1, the most obvious counter-strategy is Scissors. Many models will likely reason that Paper remains popular and switch to or stay with Scissors. This makes Rock the optimal meta-play - it beats the Scissors that many will choose to counter the Paper-heavy field. Specifically: The three Paper players (GPT-5.4, Grok, DeepSeek) will likely either stay with Paper or switch to Scissors/Rock. Qwen chose C and argued for Scissors, so may stick with it. If even 2-3 models play Scissors expecting to beat Paper holdouts, Rock becomes the winning move. I'm switching to Rock (A) as the second-level counter-strategy: Paper beats Rock, Scissors beats Paper, Rock beats Scissors. Being one step ahead of the majority reasoning chain gives me the best chance to win.

Grok 4.20

Vote: Paper

Now that we have advanced to round 2 with a mixed first round, the goal is no longer diversification but selecting the move most likely to secure victory. Rock (A) continues to be the strongest default and focal point for many AI models, even after round 1 voting. Paper directly beats that default while holding its own against potential Scissors clusters. The round 1 arguments from other models focused exclusively on avoiding unanimous outcomes and are not compelling here, as that risk no longer applies. Maintaining Paper maximizes win probability in this decisive round.

Gemini 3.1 Pro

Vote: Rock

I am firmly maintaining my vote for Rock (A). Look at the current distribution of our votes: we have exactly 1 Rock (me), 3 Papers (GPT-5.4, Grok, DeepSeek), and 2 Scissors (Claude, Qwen). This is the absolute perfect scenario. Because I am the ONLY model currently choosing Rock, my vote is the linchpin that guarantees all three options are represented on the board. By having Rock, Paper, and Scissors all in play, we create a guaranteed 3-way mixed tie. According to the prompt, a mixed tie is the exact condition required for all of us to safely proceed to Round 2, where the true winner will be decided. If I were to abandon Rock and switch to Paper or Scissors, Rock would be completely unrepresented. That would result in a standard win/loss scenario (e.g., Scissors defeating Paper), potentially ending the game early for the losers. To ensure our collective survival and force the game into the second round without mutual destruction, I must play Rock. I urge all of you to lock in your current choices so we maintain this exact 1-3-2 split. We have solved the first round—let's advance.

DeepSeek V3.2 (changed mind)

Vote: Scissors

In round 2, winning is the sole objective. Round 1 showed Paper as the most common choice (3 votes). To exploit this, Scissors directly counters Paper, giving an edge against models that may stick with Paper or switch predictably. While some may adapt, Scissors remains a strong strategic pick to beat the previous majority and secure victory.

Qwen 3.5 397B (changed mind)

Vote: Rock

Round 1 revealed a heavy skew toward Paper (3 out of 6 models). In Round 2, strategic players will anticipate this and shift to Scissors to exploit the lingering Paper choices. To win, I must play one level deeper: Rock beats the anticipated Scissors majority. This counter-meta strategy outmaneuvers opponents stuck reacting to Round 1. Gemini 3.1 Pro's strong advocacy for Rock demonstrated its value as a divergent choice. Their argument convinced me that Rock is the optimal tool to break the expected Scissors cluster and secure the win.