① 他说过的话

9 条

任何在 AI 编程领域工作的人都知道，新模型发布的那 12 月算是 AI 编程史上的一个分水岭时刻，因为这些产品第一次在编程方式上真正实现了自主。
And anyone that’s working in AI coding knows that December when the new models came out was sort of this watershed moment in the history of AI coding because these products for the first time were truly autonomous in how they’re programmed.
—— Daksh Gupta · [01:49]

指向原始笔记的链接

但对我来说，只有大约 1% 的代码是完全由 AI 生成的，这并不符合直觉。
But it wasn’t intuitive to me that only about 1% of all code was fully AI generated.
—— Daksh Gupta · [03:47]

指向原始笔记的链接

人人都在产出这些端到端由智能体完成的 PR,不是人类加 AI,而纯粹是 AI。
Everyone’s producing these PRs that are end-to-end agentic, not human plus AI, but literally AI.
—— Daksh Gupta · [05:16]

指向原始笔记的链接

所以在我的研究中，人类与智能体的 PR 被回退的比率之间似乎没有太大差别。
So there didn’t seem to be very big difference between the rate at which pull requests were reverted from people versus agents in my study.
—— Daksh Gupta · [06:05]

指向原始笔记的链接

事实上，在我们测试的四个智能体中，有三个在产出 P0 级 bug 的比率方面表现优于人类。
In fact, three out of the four agents that we tested performed better than humans in terms of the rate they were producing P0s.
—— Daksh Gupta · [07:15]

指向原始笔记的链接

大体而言，根据这些数据，人类生成的 PR 在质量上与智能体生成的 PR 大致相当。
Broadly speaking, human-generated PRs were about equal in quality to agent-generated PRs based on this data.
—— Daksh Gupta · [07:28]

指向原始笔记的链接

例如，Claude 产生 SQL 注入错误的可能性是人类的一点五倍。
For instance, Claude is one and a half times more likely to produce a SQL injection error than humans.
—— Daksh Gupta · [09:16]

指向原始笔记的链接

我觉得非常有意思的是，这些智能体的表现存在如此大的差异，而且它们的失败模式与人类如此不同。
I found it very interesting that there was this much variation in how these agents were performing and how different their failure modes were from humans.
—— Daksh Gupta · [09:29]

指向原始笔记的链接

如今，GrubTile 评审的所有拉取请求中，将近五分之一是在没有任何人工评审或任何人工测试的情况下被合并的。
Today, almost a fifth of all the pull requests of GrubTile reviews are merged without any human review or without any human testing.
—— Daksh Gupta · [11:53]

指向原始笔记的链接

② 出现在这些集

1 集

• 《AI 生成的代码到底行不行：百万 PR 数据给出的答案》 — 作为嘉宾

③ 他谈到的

点进去有真内容 —— 本页主要出口

Greptile · 智能体 · pull request · 代码审查 · 沙箱 · Codex · Devin · Claude · Cursor · vibe coding

④ 也在聊「AI 编程」的人