AMAZINGINDEX.COM 日报快照
50.4
VOL. 2026.07
2026.07.08
← 返回 2026.07.08 日报
日报快照 · Daily Snapshot
NO. 015

LLM编程助手:聪明但笨拙

#ARTICLE HackerNews 2026.07.08
推荐指数 33.0 NO. 015 · 2026.07.08
发布2026/07/07Score53Comments28

开发者实测Claude Fable辅助编码时发现,模型虽能审查代码、提交修复,却会反复犯低级错误如误提交build目录。LLM的非确定性本质使其在需要精确性的任务上持续不可靠,与确定性工具形成鲜明对比。

Karpathy 'automating themselves away' 的预言被过度浪漫化了。当前LLM编程助手的真正瓶颈不是能力上限,而是错误模式不可预测——它能通过复杂代码审查,却在build目录这种规则明确的事情上反复翻车。

这暴露了一个被忽视的产品设计问题:LLM适合'生成-审查'循环中的人类兜底场景,而非需要100%可靠性的自动化流水线。Ragel这类确定性工具在特定领域仍不可替代,混合架构(LLM+形式化工具)可能是更务实的路径。

如果你在用AI辅助开发,建议把LLM输出当作'高级草稿'而非可自动执行的指令,关键路径上保留确定性校验层。

意见分歧 26 条评论

核心争论:LLM应否用于确定性任务:效率提升 vs 本质错配

lubujackson

Makes sense, I have had the biggest wins with AI by attacking nondeterminism whenever possible. BTW, you should probably fix the Beagle link on your homepage: https://replicated.live/beagle/

gritzko

Thanks, fixed. The runtime[1] and the scripts[2] are the practical ones. I am separating the old repo[3] into submodules since submodule recursion became smooth in Beagle. [1]: https://github.com/gritzko/jab [2]: https://github.com/gritzko/beagle-ext [3]: http

inspectorSlap

I find some of the most interesting, and catastrophic failures in my agent fine-tuning come from the clamping down of non-determinism. It is totally the correct approach, but must be handled delicately. The non-deterministic core remains, but now under bimodal pressure.

替代方案: 确定性脚本形式化方法证明正确性人类流程
查看原文 →