PILOT框架：让智能体在任务运行中实时自我改进
Most agent self-improvement happens too late: the system learns from a run only…
做长时程 Agent 的同学必看，PILOT 把自我改进从离线搬到了运行中，实测通过率和 token 效率都有显著提升，值得照着思路改造自己的训练与推理链路。
Most agent self-improvement happens too late: the system learns from a run only after that run is already over.
大多数智能体自我改进发生得太晚：系统只有在一次运行结束后才能从该次运行中学习。
This new paper, moves that learning inside the run with PILOT, the harness.
这篇新论文通过PILOT（即“驾驭器”）将学习过程移入运行之中。
A worker handles the task while a separate supervisor watches its trajectory, can steer or abort it, and writes useful procedures or failure modes into persistent skills and memory.
一个工作器处理任务，而一个独立的监督器观察其轨迹，可以引导或中止它，并将有用的程序或失败模式写入持久的技能和记忆中。
That division lets the worker focus on execution while the supervisor keeps a cleaner view of the goal, progress, and drift.
这种分工让工作器专注于执行，而监督器则能更清晰地看到目标、进展和偏差。
Using the same frozen backbone for every system, PILOT ranked 1st in 5 of 6 model-benchmark configurations. Across 20 Terminal-Bench 2.0 self-improvement iterations, its best observed pass rate improved by 14.6 percentage points with GLM-5.1 and 12.4 with Kimi-K2.6. Mean output tokens per evaluated task fell 42.9% and 47.4%, respectively.
使用相同的冻结骨干网络，PILOT在6个模型基准配置中的5个中排名第一。在20次Terminal-Bench 2.0自我改进迭代中，其最佳观察通过率在GLM-5.1上提高了14.6个百分点，在Kimi-K2.6上提高了12.4个百分点。每个评估任务的平均输出令牌分别下降了42.9%和47.4%。
For long-horizon agents, that changes the self-improvement loop: a lesson does not have to wait for the next rollout. It can recover the current attempt, get tested immediately, and become part of the system used by later workers.
对于长时程智能体，这改变了自我改进循环：教训不必等到下一次滚动。它可以恢复当前尝试，立即得到测试，并成为后续工作器所用系统的一部分。
– arxiv. org/abs/2608.26530
– arxiv. org/abs/2608.26530
Title: "PILOT in the Loop: Live Self-Improvement for Long-Horizon Agents"
标题：“循环中的PILOT：长时程智能体的实时自我改进”
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力