@ylecun

–

18 November 2023 15h54

Another open source LLM from France.
They keep piling up.

Self-Supervised Learning for Planning: A Decades-Long Research Journey

By

–

18 November 2023 11h18

I've been advocating for architectures that can plan using a world models trained Self-Supervised Learning for a very long time.
E.g. my NIPS 2016 keynote (below, 25 minutes in). I've been working on SSL from video since 2014. But, unlike others:
1. I've always said that deep

AI Systems Save Lives Through Emergency Braking Medical Imaging

By

–

18 November 2023 10h00

No.
The overall number of deaths caused by AI is very negative: deep-learning based automatic emergency braking systems and medical image processing systems have been saving lots of lives.
The number of AI-fueled deaths is close to zero.

Conceptual Progress in AI Beyond Compute and Data

By

–

18 November 2023 9h02

Still lots of conceptual progress to be made in AI.
This is a scientific question.
Not merely a question of more compute and more data.

Governments Must Distinguish Expert Voices From Uninformed Opinions

By

–

18 November 2023 7h24

Governments should be careful to listen to people who know what they are talking about, and be skeptical of people who don't. https://t.co/Yqk92yU0X6
— Yann LeCun (@ylecun) 18 novembre 2023

Governments should be careful to listen to people who know what they are talking about, and be skeptical of people who don't.

Science Accelerates Through Collaboration and Open Communication

By

–

18 November 2023 7h19

Science is not a zero-sum game.
With more scientists, research accelerates.
But no one has a monopoly on good ideas.
So, with and more and faster communication, research accelerates further.