标签: AI应用

ClickHouse 和 Hud 为 AI 生成的软件构建运行时反馈循环

ClickHouse 和 Hud 为 AI 生成的软件构建运行时反馈循环

Hud与ClickHouse联合推出集成,将函数级运行时行为数据与ClickStack基础设施可观测性打通,为AI生成代码构建从发布前评估、上线后验证到异常自动回滚的反馈闭环。其背景是AI生成的代码占比已超过四成,生产环境数据正从“事后排查工具”变成“开发决策依据”。

Google真的想在AI领域获胜吗?

Google真的想在AI领域获胜吗?

Google上周对DeepMind进行了重大重组,首席科学家Jeff Dean离职创业,DeepMind CEO Demis Hassabis转任董事长专注长期研究。这一调整被市场视为Google在AI前沿竞争掉队的信号,但其Search等业务仍是巨大的安全垫。

Graph engineering is the new loop engineering which is the new harness engineering which is the new context engineering which is the new prompt engineering. Hope this clarifies any confusion.

Graph engineering is the new loop engineering which is the new harness engineering which is the new context engineering which is the new prompt engineering. Hope this clarifies any confusion.

风投机构First Mark合伙人Matt Turck用一句话调侃AI行业术语的快速更迭——图工程、循环工程、工具链工程、上下文工程、提示工程,本质都是同一件事的层层换皮。这条推文之所以引发共鸣,是因为它戳中了AI应用开发中一个真实的困惑:概念越来越多,但解决问题的思路其实一脉相承。

Running GBrain with Codex or Claude Code should be a *separate agent* and I wouldn’t expect you to run it in your main coding agent. Think of it more like a personal AI version of ChatGPT or Claude that has its own gi…

Running GBrain with Codex or Claude Code should be a *separate agent* and I wouldn't expect you to run it in your main coding agent. Think of it more like a personal AI version of ChatGPT or Claude that has its own gi...

Y Combinator 总裁 Garry Tan 发布了个人 AI 代理 GBrain 的最新版本,明确建议用户把它作为独立代理运行,而不是塞进日常编码工具里。这个“有独立 git 仓库做记忆”的定位,正在把个人 AI 从聊天工具推向真正能长期积累用户数据的私人助手。

Sources: legal AI startup Legora is in early talks to raise funds at a $10B+ valuation, up from $5.6B four months ago; its ARR rose 50% QoQ to $150M in Q2 (Financial Times)

Sources: legal AI startup Legora is in early talks to raise funds at a $10B+ valuation, up from $5.6B four months ago; its ARR rose 50% QoQ to $150M in Q2 (Financial Times)

法律AI初创公司Legora被曝正以超过100亿美元的估值洽谈新一轮融资,较四个月前的56亿美元接近翻倍;其Q2年度经常性收入(ARR)达1.5亿美元,环比增长50%。法律垂直领域的高客单价与强合规需求,正让资本市场给出远超通用AI应用的增长溢价。

新兴多智能体系统的模式与问题

新兴多智能体系统的模式与问题

Anthropic 用真实漏洞挖掘实验比较了“多智能体协作”与“独立并行”两种方式,发现协作能挖出更多漏洞,但 token 成本更高、结果更难预测;多智能体系统在规模化的路上,协调问题比模型能力更值得警惕。