Fei-Fei Li谈Atlas模型研发历程
Fei-Fei Li亲述Atlas模型研发故事，揭秘计算限制下的突破时刻和团队决策过程。
World Labs联合创始人Fei-Fei Li和Justin Johnson分享了Atlas模型的研发故事。Li表示团队对扩展定律有完全信心，而架构选择和数据混合是关键细节。Johnson提到当前受计算能力限制，模型规模扩大、训练时间延长和芯片增加都显著提升了性能。一个突破性时刻发生在夏季，当团队看到模型成功生成视角后，五秒内决定全力构建此模型。
World Labs co-founders Dr. Fei-Fei Li and Justin Johnson on compute as the constraint, and the Slack message that convinced them to go all-in on their Atlas model: Fei-Fei: "I think [we] have total conviction about the scaling law." "I do think the exact architecture choices and data mixtures is where the devil's in the details. I watched Justin and his team going from 'we really don't know how long this is gonna take,' to 'maybe sign of life,' to 'wow, this is gonna work.'" Justin: "We're basically at the beginning, and we're basically limited by compute at this point." "During development, we trained a sequence of models, the first couple rungs of the scaling ladder. Each time we made the model bigger, each time we trained it for longer, each time we put it on more chips, it got significantly better." Fei-Fei: "Here's a little bit of an insider story... There was one day in early summer... Ben and Justin feed [a smaller model] into the viewpoint generation... Remember that famous garden table from the NeRF paper?... Overnight we all saw the Slack from Ben that our camera flew under the table." "That morning, the three of us looked at each other in the eyes and said, 'That's it. We're gonna build this.' We made a decision within five seconds. No one has ever seen this result." @drfeifei @jcjohnss @BenMildenhall @martin_casado Your browser does not support the video tag. 🔗 View on Twitter Fei-Fei Li @drfeifei x.com/i/article/2104… 🔗 View Quoted Tweet 💬 6 🔄 5 ❤️ 21 👀 8526 📊 6 ⚡
