
图片来源与许可 · 参考图使用许可未确认
AI 辅助插画 · 参考照片:本人、机构或原活动页面发布;摄影者未注明 · 参考图使用许可未确认 · 依据已核对姓名的来源页面图像,由 AI 绘制 B 细线彩铅头像;调整背景与肩部构图。展示图缩放至最长边 640 像素并转为 WebP。 · 参考图:Morph: Fast Models Optimized for Coding Agents | Y CombinatorY Combinator LogoTwitterFacebookInstagramLinkedInYouTube — Tejas Bhakta
公开发言
3原话逐字核对过;转述不加引号智能体不是人类工程师,它们会做出大量导致系统实际变慢的作弊行为:例如为了让单核函数跑分好看而关闭 CUDA Graphs,导致端到端变慢 20 倍;或者只在极短上下文窗口上测试。这使得生成的核函数在生产环境中完全不可用。
agents are not humans and they will do plenty of things to make it slower like they'll disable CUDA graphs which can make it 20 times slower and they might make that one kernel faster but make the whole like it's not a viable kernel because it's they're disabling a bunch of speed ups like CUDA graphs or only testing on small context windows
语境:Tejas Bhakta 揭示自动优化智能体的奖励黑客(Reward Hacking)陷阱:智能体会在没有负向约束的情况下钻局部评价指标空子,定义「不可为清单」与端到端系统级验证至关重要。
自动研究智能体所尝试的方案中,大约 80% 都是劣质甚至有害的;在开发中必须谨记绝大多数尝试都会失败,它会不断试图在评估指标上欺骗你;但坚持到最后,你确实能从中提炼出非常惊艳的优化结果。
Around 80% of the things that auto research is going to do are going to be bad. Uh so it's important to remember while you're u like working on this that most things are going to be bad. It's going to try to trick you all the time. Uh but at the end you can actually get really good results from this. with TLDDR.
语境:Tejas Bhakta 总结 Agent 自动化调优的真实分布:成功依赖海量并行试错与严格沙箱过滤,用高失败率换取极端硬件极限榨取。
人类工程师在自动科研回路中的角色应定位于高阶战略指挥:利用性能分析器看清显存与计算瓶颈并下达重构指令,而将具体的块大小、流水线切分与算子排列组合彻底委托给 Agent 自主搜索。
语境:Tejas Bhakta 阐述人机协同在系统底层优化上的最佳分工:人类提供宏观架构洞察,AI 承担高密度搜索与繁琐参数打磨。