Dyna Robotics发布Dyna-2：百万小时人类视频预训练提升机器人操作
Maybe robots don’t only need more robot data.
Maybe robots don’t only need more robot data.
也许机器人不仅仅需要更多的机器人数据。
Humans already provide a gigantic record of interacting with the physical world. Why not learn from that ?
人类已经提供了与物理世界互动的巨大记录。为什么不从中学习呢？
Dyna Robotics just proved robot prediction improves monotonically as human video pre-training reaches 1Mn hours.
Dyna Robotics刚刚证明，随着人类视频预训练达到100万小时，机器人的预测能力单调提升。
They just launched Dyna-2, a world-action model pre-trained using 1 million hours of human video. The key breakthrough is a human-to-robot transfer scaling law.
他们刚刚推出了Dyna-2，一个使用100万小时人类视频预训练的世界动作模型。关键突破是人与机器人迁移的缩放定律。
More human video went in, better manipulation came out. And somehow, that improvement carried over to robots the model had never trained on.
更多的人类视频输入，带来了更好的操作能力。而且，这种改进竟然也迁移到了模型从未训练过的机器人上。
At unseen customer sites, Dyna reports an 87% production pass rate for Dyna-2 versus 46% for Dyna-1 at identical post-training budgets.
在未见过的客户现场，Dyna报告Dyna-2的生产通过率为87%，而在相同的后训练预算下，Dyna-1仅为46%。
So there is now an obvious next experiment: keep scaling.
所以现在有一个明显的下一步实验：继续扩大规模。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力