Meta为了新模型，居然让几千人停下手头工作去搞人工标注，可见高质量数据比算力还稀缺。想了解前沿模型训练的内幕？听听CTO怎么说。
Meta CTO确认过去一年CEO扎克伯格（Zuck）必须进入“创始人模式”推动项目，并感谢Scale AI创始人Alexandr Wang加入团队。为构建最新模型，Meta执行紧急转型，抽调数千名员工脱离常规职责，生成“专家轨迹”——即人工指导的编码示例，用于模型后训练。这表明前沿模型的瓶颈不仅是算力，更是高保真、人工验证的强化数据。该信息来自CapexAndChill的X平台播客片段。
源：https://t.co/Vosqbi8XfV
源： x.com/CapexAndChill/… CapexAndChill @CapexAndChill Meta CTO confirmed that over the last year Zuck had to go founder mode. He also really appreciated Alexandr Wang joining the team last year. To build their latest model, Meta executed an emergency pivot, pulling thousands of employees off their normal duties to generate expert traces, which are hand-guided coding examples required to post-train the model. This confirms that the bottleneck for frontier models is not just raw compute, but high-fidelity, human-verified reinforcement data. Good listen. $META Your browser does not support the video tag. 🔗 View on Twitter 🔗 View Quoted Tweet 💬 0 🔄 0 ❤️ 0 👀 211 ⚡