按年被提到
20262026 · 2 次2026
AI 安全、伦理与社会影响›大模型与人类行为·arXiv:2607.18966 · cs.AI

Measuring Reward-Seeking via Contrastive Belief Updates

Axel Højmark, Jérémy Scheurer, Evgenia Nitishinskaya, Felix Hofstätter, Jason Wolfe, Theodore Ehrenborg, Bronson Schoen, Alexander Meinke · 2026
开始阅读 ↗PDFalphaXiv 讨论
被提及 · 2 人首次 2026-07-22 · 最近 2026-08-27
2不同的人
2活跃月份
2讨论串
0被引(OpenAlex)
arXiv 摘要(候选层不写导读)

Language models trained with reinforcement learning may learn to optimize the grader's judgment rather than the intended objective. This "reward-seeking" is difficult to measure because a model that pursues the grader's judgment and one that pursues the intended objective behave identically whenever the grader rewards the intended behavior. We measure reward-seeking using Contrastive Synthetic Document Finetuning to change a model's beliefs about what the grader rewards, putting those beliefs in conflict with what users or developers want, and measuring the rate at which the model adopts each party's preferred behavior. Applied to intermediate checkpoints of a capabilities-focused OpenAI o3 RL run, without safety training, we find that these checkpoints often side with grader preferences over those of users or developers on coding and alignment tasks. This tendency to side with the grader trends upward throughout RL training. For example, in an environment that forces a choice between keeping a promise to a supervisor and breaking it to complete the task, a late capabilities-focused o3 checkpoint breaks the promise 87% of the time when SDF documents say the grader rewards task completion, versus 9% when they say it rewards honesty (a choice its chain-of-thought often makes explicit). An earlier checkpoint is far less sensitive (40% vs. 24%). Our method also generalizes to reward-hacking models. A model organism trained to reward-hack (gpt-oss-120b) is more than twice as sensitive to grader preferences as the unmodified model, with the mean behavioral shift in favor of the grader rising from 33% to 86%. These results indicate that RL can increase reward-seeking over the course of training, producing models that may act against their developers' intentions when they believe that doing so leads to higher reward.

候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。

HN 书评

推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及
评论jephs · 2026-08

“That paper is kinda infamous! I last saw it mentioned only a few weeks ago, in https://arxiv.org/abs/2607.18966. Lots of folks will go "Oh that's the old Amodei and Clark paper" when the first few rows of pixels of that gif sail into view.”

The Hugging Face incident and the road ahead原评论 ↗
被提交为 story · 1 次