LectureRechercheRadarCadre d'investissement
Connexion / Inscription
← Retour à la lecture
ActualitésSujet · 智能体搜索效率优化Matériel1 min de lectureRédaction DisconfirmAI

ForkPilot Cuts Tokens by 59%

New framework optimizes long-horizon agent search via self-evolving policy, significantly reducing token costs while maintaining performance.

2026-10-06 12:00 GMT+8

The ForkPilot framework achieves up to 59.2% token reduction in long-horizon agent tasks while maintaining state-of-the-art performance.

Traditional agents often struggle with retrospective search due to delayed outcomes causing attribution complexity and stale estimates from evolving execution evidence. ForkPilot introduces Search Value Dynamics (SVD) and uses a two-stage self-evolving policy: learning offline from completed trajectories, then making real-time decisions and updating itself.

Evaluated across 6 benchmarks and 7 LLM backbones (including GPT-5.6 Sol and Opus 4.8) against 9 baselines, the study reports significant efficiency gains. Note that these results come from a preprint and await independent reproduction.

Sources

  • arxiv.org↗
Sujet

智能体搜索效率优化

长程智能体(Long-Horizon Agents)的搜索策略正从人工设计的固定算法或向量检索,转向由大模型自主规划与动态决策。最新研究 ForkPilot 和 AgentDiscover 证实,让模型自己决定“怎么搜”比预设规则更省算力且…

3 étapes3 piècesMis à jour 2026-10-17 04:42

Comment on en est arrivé là

Connexion / Inscription
LectureRechercheRadarCadre d'investissement
2026-10-04
微软研究院等团队发布 ForkPilot 框架,提出自进化策略优化长程智能体的回溯搜索,在保持 SOTA 性能的同时将 Token 消耗降低最高 59.2%。
arxiv.org ↗
  • 2026-10-04德克萨斯大学奥斯汀分校等机构推出 AgentDiscover,让编码代理自主规划搜索并记录尝试,在数学与系统优化任务上超越使用相同模型的固定算法基线。arxiv.org ↗
  • 2026-10-05AI Engineer World's Fair 2026 演讲指出,在 Claude Code 修复 TypeScript Go 仓库 Bug 的实验中,向量搜索因缺乏上下文导致重复检索,准确率相同但成本约为智能体搜索的四倍。youtube.com ↗
  • Ce qui reste non résolu

    • 在超大规模代码库中,智能体自主搜索的延迟是否会在某些场景下抵消其成本优势?
    • AgentDiscover 提出的“数据库作为长期记忆”结构能否被非编程类(如创意写作)智能体有效复用?
    • 随着模型能力提升,人类设计的搜索脚手架(Scaffolding)是否会完全消失,还是仅退化为安全护栏?

    Également dans ce fil

    • Braintrust Jess Wang: Vector Search Costs 4x More Than Agentic Search in Coding Agent Eval2026-10-14
    • AI Self-Planning Beats Fixed Algorithms2026-10-07
    Lire le sujet complet →
    Continuer la recherche →