PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents
- ID: 3e15eb79
- 原文链接: https://arxiv.org/abs/2608.04003
- PDF: https://arxiv.org/pdf/2608.04003v1
- 作者: Shuhan Xue, Zixin Ding, Yichen Shen, Yinjie Wang, Zhenfei Yin, Yingcheng Wu, Yuxin Chen, Mengdi Wang, Ling Yang
- 日期: 2026-08-04
- 更新: 2026-08-04
- 分类: agents
- 来源类型: paper
- 标签: personal-agents, memory, recursive-self-improvement, benchmark
- 质量评分: 4/5
- 抓取时间: 2026-08-06T04:19:06Z
中文导读
PAST-Bench 把个人 AI Agent 的长期偏好任务历史工具例程和技能保留作为 recursive self-improvement 的可测场景；评测通过有序新任务序列隔离 retained experience 是否真的改善后续行为，而不是只看单次任务完成率
为什么值得关注
个人 Agent 的记忆与技能积累需要按跨会话收益来评测
PAST-Bench 把个人 AI Agent 的长期偏好任务历史工具例程和技能保留作为 recursive self-improvement 的可测场景；评测通过有序新任务序列隔离 retained experience 是否真的改善后续行为，而不是只看单次任务完成率
关键信息
- 论文标题：PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents
- 作者：Shuhan Xue, Zixin Ding, Yichen Shen, Yinjie Wang, Zhenfei Yin, Yingcheng Wu, Yuxin Chen, Mengdi Wang, Ling Yang
- arXiv：https://arxiv.org/abs/2608.04003
- 发布时间：2026-08-04
- arXiv 分类：cs.CL
- 关联标签：personal-agents, memory, recursive-self-improvement, benchmark
English Abstract
Recursive self-improvement requires agents to turn accumulated experience into better future behavior. Personal AI agents offer a concrete setting for studying this capability because they retain preferences, task histories, tool routines, and learned skills across sessions. Yet whether retained experience actually improves them over time has not been systematically tested. We introduce PAST-Bench, a benchmark designed to isolate this question. Each agent runs through ordered sequences of fresh-session tasks under matched conditions that turn retained experience on and off. It spans 26 scenarios and 204 episodes across memory, procedural reuse, information gathering, and update. We report both later-task gains and whether those gains follow the intended save, retrieve, and update pathway. Across seven base models and four agent frameworks, improvement is real but uneven across capabilities. Agents with the same headline gain can differ markedly in whether that gain is supported by evidence of the intended pathway. Guided by these findings, we develop Hermes+, which extends Hermes with five targeted interventions across stages of the agent loop. Hermes+ raises the average gain from retained experience and provides clearer pathway evidence, with its strongest improvement on tasks requiring outdated state to be replaced, although the effect remains capability- and model-dependent. Together, PAST-Bench and Hermes+ provide an evaluation and diagnostic foundation for studying how persistent agents can progress from retaining experience to systematically improving through it. Code: https://github.com/Gen-Verse/PAST-Bench
English Summary
Recursive self-improvement requires agents to turn accumulated experience into better future behavior. Personal AI agents offer a concrete setting for studying this capability because they retain preferences, task histories, tool routines, and learned skills across sessions. Yet whether retained experience actually improves them over time has not been systematically tested. We introduce PAST-Bench, a benchmark designed to isolate this question. Each agent runs through ordered sequences of fresh-session tasks under matched conditions that turn retained experience on and off. It spans 26 scenarios and 204 episodes across memory, procedural reuse, information gathering, and update. We report both later-task gains and whether those gains follow the intended save, retrieve, and update pathway....
Obsidian Notes
- 内容由
opencli arxiv paper拉取 arXiv 元数据与摘要生成。 - 中文导读与价值判断均锚定在条目已有摘要、论文摘要、作者、日期与分类信息上；未补充论文摘要之外的实验细节。