Skip to content
TheChatArchiveThe Chat Archive
Geminigemini-2.5-flash· GoogleHugging Face LMArena Human Preference 140K

This is a test

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Claudeclaude-sonnet-4-20250514’s answer.
  1. User

    This is a test

  2. Gemini Assistant

    Okay, I'm ready! How can I assist you with your test?

  3. User

    I would like to know the optimal strategy in rock paper scissors against a player who is randomly choosing between 3 strategies - 1) Random selection from all three options, 2) Playing whatever would defeat what the opponent last threw (starting with random) and 3) randomly selecting between rock and scissors, with no chance of paper.

  4. Gemini Assistant

    This is a fantastic Rock Paper Scissors problem! The key is that the opponent is randomly choosing between the three strategies each round, not picking one strategy at the start and sticking to it. This greatly impacts how you can exploit them.

    Let's break down each of the opponent's strategies and then analyze the optimal counter.

    Opponent's Strategies:

    1. S1: Pure Random (1/3 R, 1/3 P, 1/3 S)

      • Description: The opponent randomly picks Rock, Paper, or Scissors with equal probability.
      • Optimal Counter (if this was the only strategy): Any strategy of your own (Rock, Paper, Scissors, or even random) would yield a 1/3 win rate, 1/3 loss rate, 1/3 tie rate. You cannot exploit pure randomness.
    2. S2: Counter-Opponent's Last Move (starting with random first move)

      • Description: If you threw Rock last, they will throw Paper. If you threw Paper last, they will throw Scissors. If you threw Scissors last, they will throw Rock.
      • Optimal Counter (if this was the only strategy, and you played consistently): If you consistently play the same move (e.g., always Rock), S2 will consistently play the counter to that (e.g., always Paper), and you will consistently lose.
        • However, S2 relies on your last move. If you play randomly, S2's move becomes unpredictable relative to your next move.
        • If you knew they were S2, you would play the move that beats what they will play to counter your last move.
          • If your last move was Rock, S2 plays Paper. You should play Scissors.
          • If your last move was Paper, S2 plays Scissors. You should play Rock.
          • If your last move was Scissors, S2 plays Rock. You should play Paper.
        • This pattern means you want to play the move that would have tied S2's previous move. This allows you to win almost every time after the first round (where their move is random).
    3. S3: Rock/Scissors Only (1/2 R, 1/2 S, 0 P)

      • Description: The opponent will never play Paper. They randomly choose between Rock and Scissors.
      • Optimal Counter (if this was the only strategy):
        • If you play Rock: 50% Tie (vs R), 50% Win (vs S). (50% W, 0% L, 50% T)
        • If you play Paper: 50% Win (vs R), 50% Loss (vs S). (50% W, 50% L, 0% T)
        • If you play Scissors: 50% Loss (vs R), 50% Tie (vs S). (0% W, 50% L, 50% T)
        • Clearly, playing Paper maximizes your wins against S3, giving you a 50% win rate.

    Developing the Optimal Strategy

    We need a single strategy (which can be a mixed strategy, i.e., playing R, P, S with certain probabilities) that performs best on average against the opponent's random choice of these three. Let's denote our probabilities of playing Rock, Paper, and Scissors as PRP_R, PPP_P, and PSP_S respectively, where PR+PP+PS=1P_R + P_P + P_S = 1.

    Let's calculate our expected win rate (EWR) against each opponent strategy, then average them since each strategy has a 1/3 chance of being chosen.

    1. Our EWR vs S1 (Pure Random):

      • No matter what probabilities (PR,PP,PSP_R, P_P, P_S) we choose, our win rate against a purely random opponent is always 1/3.
      • EWR(vs S1) = 1/3
    2. Our EWR vs S2 (Counter Last):

      • S2 looks at our last move.
      • If our last move was Rock (happens with probability PRP_R), S2 will play Paper. Our current move is chosen with (PR,PP,PSP_R, P_P, P_S). We only win if we play Scissors (prob PSP_S).
      • If our last move was Paper (happens with probability PPP_P), S2 will play Scissors. We only win if we play Rock (prob PRP_R).
      • If our last move was Scissors (happens with probability PSP_S), S2 will play Rock. We only win if we play Paper (prob PPP_P).
      • So, our total EWR vs S2 (assuming our previous move was also chosen with our current probabilities): EWR(vs S2) = PR⋅PS+PP⋅PR+PS⋅PPP_R \cdot P_S + P_P \cdot P_R + P_S \cdot P_P
    3. Our EWR vs S3 (Rock/Scissors Only):

      • S3 plays Rock (1/2) or Scissors (1/2).
      • We win if:
        • We play Paper (prob PPP_P) AND S3 plays Rock (prob 1/2) →PP⋅(1/2)\rightarrow P_P \cdot (1/2)
        • We play Rock (prob PRP_R) AND S3 plays Scissors (prob 1/2) →PR⋅(1/2)\rightarrow P_R \cdot (1/2)
      • EWR(vs S3) = (PP/2)+(PR/2)=(PP+PR)/2(P_P/2) + (P_R/2) = (P_P + P_R)/2
      • Since PR+PP+PS=1P_R + P_P + P_S = 1, then PR+PP=1−PSP_R + P_P = 1 - P_S.
      • So, EWR(vs S3) = (1−PS)/2(1 - P_S)/2

    Total Expected Win Rate (Overall):

    Since the opponent randomly picks one of the three strategies each round (1/3 chance for each):

    Overall EWR = (1/3) * EWR(vs S1) + (1/3) * EWR(vs S2) + (1/3) * EWR(vs S3) Overall EWR = (1/3) * (1/3) + (1/3) * (PRPS+PPPR+PSPP)(P_R P_S + P_P P_R + P_S P_P) + (1/3) * (1−PS)/2(1 - P_S)/2

    Let's substitute PP=1−PR−PSP_P = 1 - P_R - P_S:

    Overall EWR = 1/9 + (1/3) * (PRPS+(1−PR−PS)PR+PS(1−PR−PS)P_R P_S + (1 - P_R - P_S) P_R + P_S (1 - P_R - P_S)) + (1/3) * (1−PS)/2(1 - P_S)/2

    This looks complicated to optimize directly. Let's think about the goal: maximize win rate.

    Notice that S3 never plays Paper. This means if we play Paper, we can only win (vs R) or lose (vs S), never tie. If we play Scissors, we can only lose (vs R) or tie (vs S), never win. This suggests we should try to maximize our use of Paper and minimize our use of Scissors. If we can avoid Scissors entirely (PS=0P_S = 0), that looks promising for S3.

    Let's test PS=0P_S = 0 (never play Scissors). This means PR+PP=1P_R + P_P = 1.

    If PS=0P_S = 0:

    • EWR(vs S1) = 1/3 (still the same)
    • EWR(vs S2) = PR⋅0+PP⋅PR+0⋅PP=PRPPP_R \cdot 0 + P_P \cdot P_R + 0 \cdot P_P = P_R P_P
    • EWR(vs S3) = (1−0)/2=1/2(1 - 0)/2 = 1/2

    Now, substitute PP=1−PRP_P = 1 - P_R into EWR(vs S2): EWR(vs S2) = PR(1−PR)P_R (1 - P_R)

    Overall EWR (if PS=0P_S = 0) = (1/3) * (1/3) + (1/3) * PR(1−PR)P_R (1 - P_R) + (1/3) * (1/2) Overall EWR = 1/9 + (1/3) * PR(1−PR)P_R (1 - P_R) + 1/6 Overall EWR = 5/18 + (1/3) * (PR−PR2P_R - P_R^2)

    To maximize this expression, we need to maximize PR−PR2P_R - P_R^2. This quadratic function is maximized when PR=1/2P_R = 1/2. If PR=1/2P_R = 1/2, then PP=1−PR=1/2P_P = 1 - P_R = 1/2.

    So, the optimal strategy is to play Rock 50% of the time, Paper 50% of the time, and never Scissors.

    Let's calculate the overall EWR for this strategy (PR=0.5,PP=0.5,PS=0P_R=0.5, P_P=0.5, P_S=0):

    1. Vs S1 (Random):

      • Our Win Rate = 1/3 (as always against true random)
    2. Vs S2 (Counter Last):

      • If our last move was Rock (0.5 prob), S2 plays Paper. Our current move is R (0.5) or P (0.5).
        • R vs P (lose)
        • P vs P (tie)
        • No wins from this branch.
      • If our last move was Paper (0.5 prob), S2 plays Scissors. Our current move is R (0.5) or P (0.5).
        • R vs S (win!) - happens with 0.5×0.5=0.250.5 \times 0.5 = 0.25 probability.
        • P vs S (lose)
      • So, our Win Rate vs S2 = 0.25 (or 1/4).
    3. Vs S3 (Rock/Scissors Only):

      • We play R (0.5 prob) or P (0.5 prob). S3 plays R (0.5 prob) or S (0.5 prob).
      • We win if:
        • We play R (0.5) AND S3 plays S (0.5) →\rightarrow Win (0.5×0.5=0.250.5 \times 0.5 = 0.25)
        • We play P (0.5) AND S3 plays R (0.5) →\rightarrow Win (0.5×0.5=0.250.5 \times 0.5 = 0.25)
      • Our Win Rate vs S3 = 0.25+0.25=0.50.25 + 0.25 = 0.5 (or 1/2).

    Overall Expected Win Rate for the Optimal Strategy:

    Overall EWR = (1/3) * (1/3) + (1/3) * (1/4) + (1/3) * (1/2) Overall EWR = 1/9 + 1/12 + 1/6 Overall EWR = 4/36 + 3/36 + 6/36 Overall EWR = 13/36

    13/36 is approximately 36.11%. This is better than the 33.33% you'd get from playing purely randomly.

    Optimal Strategy Summary

    Your optimal strategy is to never play Scissors. Instead, randomly choose between Rock and Paper with equal probability (50% for Rock, 50% for Paper).

    This strategy performs as follows:

    • Against the Pure Random player (S1), it maintains the baseline 33.3% win rate.
    • Against the "Rock/Scissors Only" player (S3), it leverages Paper's advantage to achieve a 50% win rate.
    • Against the "Counter Last" player (S2), it avoids predictable patterns for Scissors and achieves a 25% win rate, which is higher than the 0% you'd get if you were always predictable.

    The slight improvement comes from exploiting S3 without making yourself too vulnerable to S2.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.