Skip to content
Hot eventLive

D-OPCD:将智能体经验蒸馏进扩散模型权重

1 reports1 sources3 hr ago updated

Get the story

AI overview

2026年10月7日,arXiv(Artificial Intelligence,一手来源)发表研究,提出 Diffusion On-Policy Context Distillation(D-OPCD)。该方法把智能体改进后的提示词作为特权上下文,通过在线策略上下文蒸馏,将 agent harness 的能力内化到扩散模型权重中。这是目前该事件唯一一篇报道,尚未出现后续验证、复现或争议信息。

Generated from reports · updated 3 hr ago

Timeline

Follow the coverage from different angles.

Oct 7, 2026
  1. arXiv · Artificial Intelligence
    D-OPCD:用在线策略上下文蒸馏将智能体经验内化到扩散模型权重

    研究提出 Diffusion On-Policy Context Distillation(D-OPCD),把智能体改进后的提示词作为特权上下文,将 agent harness 的能力蒸馏进扩散模型权重。

Heat trend

Current heat 9·Comparable peak 10(Oct 7)·Comparable change over 24 hours –

02.557.510Oct7Oct7Oct7Oct7

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.