Microsoft proposes Fork-and-Flush to improve AI agent search
What happened
On October 13, The Decoder reported that Microsoft Research proposed Fork-and-Flush, a method that runs parallel branches and clears chat context to improve search results for long-running AI research agents. Each branch keeps its code, files and saved results, continues exploring in a fresh conversation, and the best branch is kept afterward. Across 13 tasks under the same compute budget, the method scored a normalized average of 0.78, against 0.47 for a single run and 0.54 for best-of-4, relative gains of 66.0% and 44.4% over the two baselines.
Written by AI from the coverage · updated 3 hours ago
Coverage
Follow the reports to see the story from different sides.
- The DecoderClearing chat history helps AI agents find better solutions
Microsoft研究提出Fork-and-Flush，通过并行分支与清空聊天上下文，改善长期运行AI研究智能体的搜索结果。各分支保留代码、文件和已保存结果，以新对话继续探索，随后保留最佳分支；在13项任务、相同计算预算下，归一化平均得分为0.78，单次运行为0.47，best-of-4为0.54，相对提升分别为66.0%和44.4%。
Heat over time
Heat now 9·Comparable peak 10（Oct 13 01:00）·Comparable change over 24 hours –
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.