Florent de Stael

Tech enthusiast and entrepreneur with a heart for our planet and animal advocacy.
(Photo by John Moeses Bauan on Unsplash)

[Florent de Stael @destael]

@jordibruin Did you stop MacGPT dev? I did not see any update for a while.

[Florent de Stael @destael]

@BazQuxReader login issue . Mobile access also not working

[Florent de Stael @destael]

@BlaBlaCarDaily Ok merci pour le retour. Pas de covoiturage ce matin pour moi, à une prochaine fois...

[Florent de Stael @destael]

@BlaBlaCarDaily votre application est en panne, impossible de co-voiturer ce matin. Une page Statut serait la bienvenue

[Florent de Stael @destael]

@VandelPhilippe @ProtonSupport @ProtonPrivacy @ProtonMail Same for me

[Florent de Stael @destael]

@ProtonSupport unable to access to protonmail

[David Chavalarias @chavalarias]

🚨Le Kremlin à l'Assaut de la République : Une Étude révèle les manœuvres de déstabilisation de la France
➡Étude #Politoscope avec révélations cruciales sur l'affaiblissement du front républicain à l'approche du 2nd tour des #legislatives2024
https://t.co/6TJhyGP6ZA

[Florent de Stael @destael]

It's crystal clear. Merci @ylecun

[Yann LeCun @ylecun]

* Language is low bandwidth: less than 12 bytes/second. A person can read 270 words/minutes, or 4.5 words/second, which is 12 bytes/s (assuming 2 bytes per token and 0.75 words per token). A modern LLM is typically trained with 1x10^13 two-byte tokens, which is 2x10^13 bytes. This would take about 100,000 years for a person to read (at 12 hours a day).

* Vision is much higher bandwidth: about 20MB/s. Each of the two optical nerves has 1 million nerve fibers, each carrying about 10 bytes per second. A 4 year-old child has been awake a total 16,000 hours, which translates into 1x10^15 bytes.

In other words:
- The data bandwidth of visual perception is roughly 16 million times higher than the data bandwidth of written (or spoken) language.
- In a mere 4 years, a child has seen 50 times more data than the biggest LLMs trained on all the text publicly available on the internet.

This tells us three things:
1. Yes, text is redundant, and visual signals in the optical nerves are even more redundant (despite being 100x compressed versions of the photoreceptor outputs in the retina). But redundancy in data is *precisely* what we need for Self-Supervised Learning to capture the structure of the data. The more redundancy, the better for SSL.
2. Most of human knowledge (and almost all of animal knowledge) comes from our sensory experience of the physical world. Language is the icing on the cake. We need the cake to support the icing.
3. There is *absolutely no way in hell* we will ever reach human-level AI without getting machines to learn from high-bandwidth sensory inputs, such as vision.

Yes, humans can get smart without vision, even pretty smart without vision and audition. But not without touch. Touch is pretty high bandwidth, too.

[Florent de Stael @destael]

@AFnewsroom bonjour @AFnewsroom, je ne postule pas mais j'enseigne l'IA dans une école de commerce et je voudrais organiser une interview sur le sujet, peut-on se parler en DM ?

[Florent de Stael @destael]

Blizzard employee snags a decade of WoW game time just before being laid off, will not have to pay a cent until 2033 https://t.co/wohNwewuhi

[The Debian Project @debian]

Statement about the EU Cyber Resilience Act https://t.co/Nizj5UJGQi

[Florent de Stael @destael]

Zoom change ses conditions d'utilisation et s'octroie le droit d'utiliser les données des utilisateurs pour entraîner son IA sans leur consentement, ni possibilité de s'y opposer https://t.co/eImpogxVhb via @developpez

[Florent de Stael @destael]

@CBeaune @Ecologie_Gouv @DGAC @ParisAeroport un lien pour avoir le Replay ? Vraiment curieux de suivre cet échange et les arguments de @JMJancovici pendant ce salon

[Florent de Stael @destael]

bonjour @salondubourget j'ai entendu parlé d'un table ronde « Destination : quel avenir pour l’aviation ? », avec la participation de @JMJancovici. Peut-on la voir en replay ?

[Florent de Stael @destael]

Enfin un signe de l'intelligence de #ChatGPT . 😂.

[Patrice Le Foll @plefoll]

"#ChatGPT n’est pas prêt pour répondre aux clients de la #SNCF" https://t.co/cQ4zeB2eK4

[Florent de Stael @destael]

Un beau geste @Xavier75 , soutenez la presse informatique libre

[Next.ink @nextinpact]

Next INpact à la croisée des chemins https://t.co/ZL0WUFKo0i

[Intel @intel]

Today, we lost a visionary.

Gordon Moore, thank you for everything.

[10TO11 @10to11tech]

Petit essai chez 10TO11 du nouvel logiciel de Microsoft, Designer (https://t.co/EogmzpKDjw). Avec l'IA, en tapant le titre l'illustration est générée automatiquement. Pour ceux qui connaissent Monaco, ce n'est pas tombé très loin pour un premier essai, il manque juste 10 km !

[vitalik.eth @VitalikButerin]

@coinmamba I'd recommend increasing your distance from trading/investing circles, and getting closer to the tech and application ecosystem. Learn about ZK-SNARKs, visit a meetup in Latin America, listen to All Core Devs calls and read the notes until you've memorized all the EIP numbers...