标签: 大模型

借助会话追踪与成本控制排查 AI 智能体故障

借助会话追踪与成本控制排查 AI 智能体故障

StackGen 首席工程师 Sabith K Soopy 在 8 月 4 日发布的 CNCF 博文中,分享了用会话追踪(Langfuse)和成本控制来排查 AI 智能体故障的生产经验——智能体反复调用错误工具时往往不会触发传统可用性告警,需要专门的观测手段才能定位问题。

I was not expecting things to go this way, but I think MCPs are better than CLIs for most integrations. The models have gotten much better at tool calling, we can defer tools & MCP is now stateless. If you need to…

I was not expecting things to go this way, but I think MCPs are better than CLIs for most integrations. The models have gotten much better at tool calling, we can defer tools & MCP is now stateless. If you need to...

开发者 Thariq 在 X 上提出一个与主流预期相反的观点:随着大模型工具调用能力提升、工具可以延迟加载、MCP 协议转向无状态,MCP 在多数集成场景中已经比 CLI 更合适。这值得关注,因为它触及 AI 应用接入外部工具的标准路线之争。

The Grok @bot team asked me if I had any advice for solopreneurs. I’ve only been doing this full time for three months, but my advice is to design your business so you can spend more time doing work you actually enjoy…

The Grok @bot team asked me if I had any advice for solopreneurs. I’ve only been doing this full time for three months, but my advice is to design your business so you can spend more time doing work you actually enjoy...

独立开发者 Peter Yang 在参与 Grok @bot 团队的一场直播活动时,被问到对“一人公司”(solopreneur)的建议,他给出的答案不是如何多赚钱,而是把业务设计成能让自己多花时间做真正喜欢的事,把无聊工作交给 AI bot 处理。

There’s a massive chasm between the power of AI models and the ultimate workflows that enterprises are trying to automate. This gap is the opportunity for the applied AI layer to fill. You need to connect the intellig…

There’s a massive chasm between the power of AI models and the ultimate workflows that enterprises are trying to automate. This gap is the opportunity for the applied AI layer to fill. You need to connect the intellig...

Box CEO Aaron Levie 在 X 上提出,大模型能力与企业实际要自动化的流程之间存在巨大鸿沟,这个鸿沟正是"应用 AI 层"的机会所在。他强调即使模型持续进化,这一层依然必须存在,甚至可能变得更重要。

在 LLM 时代学习编程

在 LLM 时代学习编程

Hacker News 上一场关于“LLM 时代还要不要学编程”的讨论引发关注,核心结论是:AI 让初学者能更快做出超出自身理解范围的 MVP,但“能跑”并不等于“可控”,学习方式必须从“先学会再做”变成“边做边刻意补理解”。

人类学/知识工作插件

人类学/知识工作插件

Anthropic 在 GitHub 开源了 11 个面向知识工作者的 Claude 插件,覆盖销售、法务、财务、数据分析等岗位,把模型能力封装成可定制的岗位工具箱,同时兼容 Claude Cowork 和 Claude Code。