elvis@omarsar0
40AI 编辑部评分,满分 100
2026-08-11 02:26· 36分钟前
AI 导读

DynaRobotics 推出 Dyna-2,一个基于 100 万小时人类视频预训练的世界动作模型,可同时预测未来视频与动作。首次发现人类数据在 1000 至 100 万小时四个数量级上的缩放定律,并外推至未见过的机器人数据,表明世界建模与视频数据缩放对跨本体迁移至关重要。

This is huge!

Interesting new scaling laws discovered.

Dyna-2 is a world-action model trained on over 1M hours of egocentric human video.

It jointly predicts future video and future actions, so it reasons about outcomes before acting.

Dyna RoboticsToday we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several ...

来源:elvis · x.com

elvis · @omarsar0 · X·2026-08-11 02:26·36分钟前
AI 导读

DynaRobotics 推出 Dyna-2,一个基于 100 万小时人类视频预训练的世界动作模型,可同时预测未来视频与动作。首次发现人类数据在 1000 至 100 万小时四个数量级上的缩放定律,并外推至未见过的机器人数据,表明世界建模与视频数据缩放对跨本体迁移至关重要。

This is huge!

Interesting new scaling laws discovered.

Dyna-2 is a world-action model trained on over 1M hours of egocentric human video.

It jointly predicts future video and future actions, so it reasons about outcomes before acting.

Dyna RoboticsToday we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several ...

来源:elvis· x.com