智能体
x.com ·
@omarsar0
Google 论文 VeriHarness：用 Agent 验证器改进长时程任务
Google 论文 VeriHarness 指出，采样多个 Agent rollout 并信任一致答案的做法可能掩盖共享错误，而分歧往往指向正确替代方案。该方法将同一基座模型转为 agentic verifier，分别解决分歧声明和挑战全体一致的声明。在五个长时程基准上取得最佳选择分数，配合证据修订在 Gemini 3.5 Flash 和 Claude Opus 4.8 上分别提升 6.2 和 6.4 分，并发布约 26000 条 rollout。
AI推荐理由 介绍 Google 论文 VeriHarness，指出多 rollout 一致性可能掩盖共享错误、分歧常指向正确答案，并提出将同一基座模型转为 agentic verifier 的两项职责。给出五个长时程基准上的选择分数表现，以及相对单次 rollout 的 6.2/6.4 分提升，并附 arXiv 链接和约 26000 条 rollout 数据。有具体方法、基准和可验证来源，对 Agent 验证方向价值较高。
80
浏览 4,188
点赞 86
评论 17
转发 13
正文
Banger paper from Google.
It's standard practice to sample several agent rollouts and trust the answers they agree on.
This Google paper shows that agreement can hide shared errors, while disagreement often points to the correct alternative.
VeriHarness turns the same base model into an agentic verifier with two jobs.
One resolves claims where rollouts disagree by checking workspace evidence.
The other challenges claims that every rollout agrees on and looks for requirements they all missed.
Across five long-horizon benchmarks, it gives the best selection scores among the baselines tested. With evidence-backed revision, it adds 6.2 points over a single rollout with Gemini 3.5 Flash and 6.4 points with Claude Opus 4.8.
The authors also release about 26,000 rollouts.
Paper: [arxiv.org/abs/2610.00972](https://arxiv.org/abs/2610.00972)
Chat with Paper: [academy.dair.ai/papers/verihar…](https://academy.dair.ai/papers/veriharness-scaling-agentic-verification-for-long-horizon-tasks-2610.00972)
