I find it intriguing that sora / large language models can’t infer physics from all of the video used to train them. Object permanence is tricky, let alone physics.

串文

2024-12-13 18:22

讚

回覆

轉發

作者

Aaron Carámbula
carambula

粉絲

677

串文

91+

讚

回覆

轉發

24小時粉絲增長

無資料

互動率

(讚 + 回覆 + 轉發) / 粉絲數

0.44%

回覆 (BETA)

最先回覆的內容
發文後	用戶	內容
2 小時內	Wil Chung iamwil	It’s consistent from frame to frame. Just not on a longer timescale of seconds. My guess is if had to make things more consistent on a longer timescale, it’d get it right.