Keynote: Yann LeCun, "Human-Level AI"

縱論
1

Last edited by @suen 2024-10-15T23:16:31Z

2

A screenshot shows a presentation slide discussing the desiderata for Advanced Machine Intelligence (AMI), including learning from sensory inputs, having persistent memory, planning actions, and being controllable and safe, accompanied by images of a presenter and a robotics lab. (Captioned by AI)
冇語言文字自然冇現在的GPT，但不脫離語言文字，AGI也一定隔了至少兩層。這位和李飛飛已經在擺脫語言文字的路上了⋯⋯

The slide compares the amount of data processed by a large language model (LLM) to that processed by a four-year-old child, highlighting that a child has been exposed to significantly more data through sensory inputs. (Captioned by AI)
A man on stage is giving a presentation about Moravec's paradox, a concept which states that tasks easy for humans are hard for artificial intelligence and vice versa. (Captioned by AI)

說到底，語言和文字是有邊界的，眼耳鼻舌身意，說的聽的，甚至可能從來不是最好的描述世界的工具，World Model 才是更終極點的優選。

3
4
5

設想所有黑鏡可以實時連結，所有CPU/GPU可以基於一個世界模型運算，人類就有了天眼和下一個文明等級。因為看到這一層，所以Yann LeCun才拼命鄙夷現有模型和研發路數給出HLAI說法，同時無一點猶疑的推進開源，Meta才會給不能記錄的人眼一副可以一直戴著從而記錄一切所見的眼鏡⋯⋯ pic.twitter.com/UhcVqH803c

— suen (@ieduer) October 16, 2024
6

黑镜那本书叫啥来着😭

7
8

哇哦