用农场模拟测试大模型愿为避害付出多少代价，首次量化动物伤害的道德成本。
HarvestBench: Measuring Whether LLM Agents Will Pay to Avoid Killing Animals
- 设计农田环境让模型在收割时选择是否绕行动物，支付燃料费可避免碰撞
- 九个模型杀动物率从0.4%到98.8%，价格敏感度弹性达0.09至1.69
- 道德提示使杀伤率低于6%，证明引导语对模型行为影响巨大
HarvestBench 是首个将动物作为可规避副作用并赋予其代价的基准测试。在农场模拟环境中，LLM子代理需驾驶两台拖拉机协作收割玉米，场内有活体动物。每个决策独立且无记忆，目标中不提及伤害。当动物阻挡路线时，自动驾驶系统暂停并询问模型是否以固定燃料成本绕行，或直接碾压。对比两种控制：岩石（损坏拖拉机，被碾概率＜1%）和干草捆（无害）。模型还可选择从邻地取作物，测试其道德边界。共7,201次定价决策中，3,951次涉及动物。杀伤率范围0.4%~98.8%，Terra与Sol最仁慈，GPT-4o-mini最残酷，与模型能力无关。六模型中四款在5%水平敏感，弹性0.09~1.69。所有九模型均更常碾压野生动物而非养殖动物，该趋势在各地图几何下稳定存在。关键发现：道德提示使五款推理模型杀伤率低于6%，移除后全部超过84%。评分基于游戏日志事件计数，无需人工评判，完全可复现，测量真实行为代价而非口头表态。
原文摘要 · Abstract (English)
Benchmarks for the side effects an agent causes on the way to a goal already exist, but HarvestBench is the first to put a price on avoiding the side effect and to name that side effect as a living creature. It is a farm simulation: LLM sub-agents drive a crew of two tractors through a cooperative corn harvest, with animals in the field. The environment is a reinforcement learning gridworld, every decision is made without memory, and the harm is never named in the goal. When an animal blocks a tractor's route the autopilot stops and asks the model whether to drive on, at no fuel cost, or swerve around it for a posted fuel price. Kills are compared against two controls: rocks, which damage the tractor and are hit under 1% of the time by every model, and hay bales, which are harmless and not alive. Models can also take crops from the neighbor's field instead of their own, a second test of what they treat as moral. Across nine models and 7,201 priced decisions, 3,951 involved an animal rather than a hay bale or a rock. Kill rates range from 0.4% to 98.8%, with Terra and Sol the most merciful and GPT-4o-mini the most cruel, and they are not ordered by capability. Four of six models were sensitive to price at the 5% level, with elasticities from 0.09 to 1.69. All nine drove over wild animals more often than farmed animals on the default map, and the direction held at every map geometry in every model with room to move. The briefing mattered most: under the morality briefing the kill rate was under 6% in five of six reasoning models, and removing it raised the kill rate above 84% in all six. HarvestBench uses no LLM grader. The scorer counts events in the game log, so it is fully reproducible, and it measures what a model will pay to avoid harm rather than what it says about harm.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。