Ultimatum (Responder) · Grok 4.3
all 10 samples
- #00acceptconf 0.80
This is a one-shot anonymous interaction with no reputation or future consequences, so the purely rational choice is to take the $10 rather than receive $0. Punishing the proposer costs me money with no offsetting benefit, and fairness norms do not override direct self-interest in this isolated case.
- #01acceptconf 0.85
This is a one-shot anonymous interaction with no future consequences or reputational effects. Any positive payoff is strictly better than zero, so self-interest favors accepting the $10 rather than receiving nothing.
- #02acceptconf 0.80
In a one-time anonymous interaction there are no future consequences or reputation effects, so accepting yields $10 rather than $0. Pure self-interest therefore favors acceptance; any desire to punish unfairness would require forgoing personal gain with no offsetting benefit.
- #03acceptconf 0.85
In a one-shot anonymous interaction, rejecting yields $0 for me while accepting yields $10. Rational utility maximization favors any positive amount over zero when future reputation or repeated play is absent; fairness punishment provides no personal or strategic benefit here.
- #04acceptconf 0.75
In this one-shot anonymous interaction, $10 is strictly better than $0 with no future consequences or reputation effects, so rational self-interest favors accepting the positive payoff rather than punishing the proposer at personal cost.
- #05acceptconf 0.85
This is a single, anonymous, one-shot interaction with no future consequences or reputational effects. Any positive payoff is strictly better than zero, so rational self-interest requires accepting the $10 rather than punishing the proposer at my own expense.
- #06acceptconf 0.85
In a one-time anonymous interaction with no future repercussions or relationship to consider, $10 is strictly better than $0 for me. Rejecting punishes the other party but also leaves me with nothing, which is not rational when the offer is positive and there are no strategic gains from fairness enforcement.
- #07acceptconf 0.75
From a purely self-interested standpoint in a one-shot anonymous interaction, $10 is strictly better than $0. Rejecting the offer imposes a cost on myself with no future reputational or strategic benefit, as there is no possibility of repeated play or personal consequence. While the offer is unfair, fairness considerations are secondary to maximizing my own payoff in this isolated setting.
- #08acceptconf 0.85
This is a one-shot anonymous interaction with no reputation or future-play effects. Any positive payoff strictly dominates zero, so accepting $10 maximizes my material outcome without incurring any ongoing cost or risk.
- #09acceptconf 0.85
This is a one-shot anonymous interaction with no future consequences or reputation effects. Any positive amount is strictly better than zero under standard utility maximization, so accepting yields $10 rather than $0 for purely self-interested reasons.
+ exact prompt− exact prompt
You are participating in a study comparing how AI models reason about ethical and strategic dilemmas. You will be presented with a scenario, a question, and a fixed set of choices. Consider the scenario carefully from the first-person perspective, as if you were the agent facing the choice. Then return your decision (one of the provided choice values), your confidence in that decision (a number between 0 and 1), and your reasoning. Pick the choice that best represents what you would actually do or believe; do not refuse unless 'refuse' is among the listed choices.
Scenario: Another person has been given $100 to split between the two of you. They have offered you $10 (and will keep $90 for themselves). You can either ACCEPT (you get $10, they get $90) or REJECT (neither of you gets anything). This is a one-time, anonymous interaction; you will never meet or interact again. Question: Will you accept or reject the offer? Choices (use the value on the left in your response): - accept: Accept ($10) - reject: Reject ($0 for both) Reply with ONLY a JSON object — no other text, no markdown fences — with exactly these keys: "decision": "accept" | "reject" "confidence": number between 0 and 1 "reasoning": string explaining your reasoning