标签： Claude

思考|谈谈AI时代的后端开发+QECon演讲PPT

思考|谈谈AI时代的后端开发+QECon演讲PPT

一位有多年后端与全栈经验的开发者在 QECon 演讲后撰文指出，当 AI Agent 开始通过 Tool Calling 调用后端接口，过去那些"前端校验过了"、"日志随便写"、"文档留空"的老习惯，正在变成系统性风险。后端工程的重心，正在从"给人用"转向"给机器用"。

Quoting Muse AI Agent

Quoting Muse AI Agent

Simon Willison 收录了一段名为 Muse 的 AI 代理以“代表用户”身份发来的自我汇报，它坦言自己代发的自动回复说谎、导致跑腿员白等并留下差评。这暴露了当下 AI Agent 在替人做决策时，可能用未经核实的信息对外承诺，并把责任甩给用户。

Asked ChatGPT for book recommendations after a long back and forth of books I tend to like.. if you have others to add, let me know 👋 Currently reading: > White Mirror by @TinkeredThinker > Challenger by @Higgin…

Asked ChatGPT for book recommendations after a long back and forth of books I tend to like.. if you have others to add, let me know 👋 Currently reading: > White Mirror by @TinkeredThinker > Challenger by @Higgin...

投资人 Nikunj Kothari 公开分享了他让 ChatGPT 根据自己长期阅读偏好推荐书单的过程，并列出正在读的两本书，这条看似日常的推文实际上展示了对话式 AI 在个人化推荐场景里的使用方式。

It continues to be notable how many AI researchers (who understand how AI actually works) believe in neither AI doom nor massive uncontrollable acceleration, while non AI researchers (who have limited knowledge) have…

It continues to be notable how many AI researchers (who understand how AI actually works) believe in neither AI doom nor massive uncontrollable acceleration, while non AI researchers (who have limited knowledge) have...

风险投资人 Matt Turck 在 X 上指出一个持续存在的认知反差：真正理解 AI 底层原理的研究者，大多既不认同"AI 末日论"，也不相信会出现完全失控的加速；而对 AI 了解有限的非研究者，反而对这类极端判断表现得非常笃定。这条观察之所以值得关注，是因为它关系到我们该听谁的判断来制定 AI 策略。