Labrin

用 Typeless，我时常会误以为我的网络是 3G 网络 反复失败，然后不得不自己重新打字 治疗低血压的 AI 产品出现了

One day, when we look back on the past, we will call it a turning point

分享一个吃法 康乐醋配烤干的吐司 感谢 @coolish 安利的醋，只是我有点上瘾了

拼命背梗依旧无法超越萌佬

这能剩下不少时间

100% agree 策略的核心还是机制分析，速度可能是机制的条件，大部分情况下只是为了攫取更多的收益； 策略的验证阶段还是以 notebook + ai 数据分析为主

Midjourney 依旧是 imagegen 领域最有审美的模型

眼睁睁看着 dgx spark 从 2万5 涨到现在 5万 犹豫就会败北

还需要啥 motebook

拼命背梗依然不敌天才

TLDR：做聚合器了

每天交流最多的是 AI，已经彻底离不开 Typeless 语音转文字了 但他们稳定性做得挺差... 寻找平替中

许久以来的疑惑原来不止我一人

今天让 Claude 帮我扫了一遍盘，给我震惊住了 Chrome 设备端的 AI 模型占了我 4 个 G 这个模型叫做 OptGuideOnDeviceModel ,是 Google 的 Gemini Nano 基本没啥用, 还占空间... 强烈建议让你的 claude 扫一下，用 Chrome 的企业策略禁止下载，然后再删除这个模型

An AI-native operating system should be agentic: the agent no longer has to wrestle with the system for low-level privileges—it can even control hardware drivers directly The operating system's firewall may well be the agent itself Users don't need to memorize complicated keyboard shortcuts. All an AI-native OS may really need is a single microphone button: once the machine boots, the user simply talks to it through the mic

There’s one fundamental point we need to recognize: most people aren’t particularly good at asking the right questions or articulating what they need. More often than not, they don’t even fully understand their own needs in the first place. This is simply human nature, and it applies to the vast majority of us. To me, the ultimate evolution of AI agents goes far beyond responding to prompts. With secure, permission-based access to our data and context, an agent should be truly proactive. It should understand our needs, anticipate problems, identify opportunities, and reach out to us with solutions—often before we even realize we need help. The future shouldn’t be about humans encountering problems, struggling to formulate imperfect prompts, and hoping an AI can figure out what they mean. It should be about AI understanding what matters to us and taking the initiative to help, without waiting to be asked.