Stéphane Halimi
LightNews LightNews
banner
stephanehalimi.bsky.social
Stéphane Halimi
@stephanehalimi.bsky.social
12 followers 60 following 7 posts
Co-createur de Siri depuis toujours ( confirmé par mon éditeur ).
Auteur du livre "l'intelligence artificielle et mes connaissances n'existent pas".
Ph. D 6 fois, hexa-doctor, inventeur des théorèmes de Thales et de Pythagore.
Posts Replies Media Videos
Reposted by Stéphane Halimi
scienceetonnante.com
Nouvelle vidéo ! LA CRYPTOGRAPHIE QUANTIQUE.

Je vous parle de BB84, le protocole de distribution de clé quantique qui a valu à Bennett & Brassard leur prix Turing il y a quelques semaines.

Et je dis un mot aussi des protocoles à base d'intrication !

youtu.be/V35v6HjG67U
La Cryptographie Quantique — [Le protocole BB84]
YouTube video by ScienceEtonnante
youtu.be
April 24, 2026 at 3:10 PM
Reposted by Stéphane Halimi
scienceetonnante.com
Nouvelle vidéo ! FUSION NUCLÉAIRE : Comment piéger un plasma sans le toucher ? On y parle du confinement magnétique et des principes du tokamak, de sa conception à son contrôle en temps réel.

Le lien vers la vidéo:
youtu.be/ltLAZsJELng
Fusion nucléaire : Comment piéger un plasma sans le toucher ?
YouTube video by ScienceEtonnante
youtu.be
June 12, 2026 at 3:06 PM
Reposted by Stéphane Halimi
ricardsole.bsky.social
Why is reading so fundamental to the human mind? In this fascinating paper, @standehaene.bsky.social & Co show how literacy reshapes and connects the brain’s visual and language systems. A timely reminder of what education could lose if LLMs replace reading & writing.
www.unicog.org/publications...
July 19, 2026 at 1:00 PM
Reposted by Stéphane Halimi
yann-lecun.bsky.social
Auto-Regressive LLMs (auto-encoders with causal transformer architectures) and BERT-style models (denoising auto-encoders with transformer archis) are smashing demonstrations of the power of self-supervised (pre-)training.
But they only work for sequences of discrete symbols: language, proteins...
eugeneyan.com Eugene Yan @eugeneyan.com · Nov 17
Also see Karparthy’s take on it
It’s hard to understand now, the Atari RL paper of 2013 and its extensions was the by far dominant meme. One single general learning algorithm discovered an optimal strategy to Breakout and so many other games. You just had to improve and scale it enough. My recollection of the memetics is that Yann LeCun was one prominent person who really didn’t care much and talked about the cake over and over again, where RL was just the final cherry on top with representation learning as the meat and supervised learning the icing, and he was conceptually exactly right about that at least with today’s stack and hindsight (pretraining = meat, SFT = icing, RLHF = cherry, ie the basic ChatGPT training pipeline). Which is fun because today he really doesn’t care much for LLMs either. (But for reasons that I tbh don’t always fully follow.)
November 24, 2024 at 9:44 PM
Reposted by Stéphane Halimi
yann-lecun.bsky.social
Aravind Srinivas (Perplexity) tells Lex all the ways in which I was right against the prevalent ideas of the time: DL, ConvNets, energy-based models, SSL, the limitation of RL, and now the limitations of auto-regressive generative models including LLMs.
Thanks Aravind!

youtu.be/mnGUfkMt9fE?...
Yann LeCun is a controversial visionary | Aravind Srinivas and Lex Fridman
YouTube video by Lex Clips
youtu.be
December 15, 2024 at 10:28 PM