
Re-author everything. Especially the opening. benoitbergeret.substack.com/…

AI founder, operator & lecturer translating complex science into clear, usable decisions for business and policy. 37 years in AI. Essays: benoitbergeret.substac

Re-author everything. Especially the opening. benoitbergeret.substack.com/…

Beyond the Chatbot, Beyond the AI Office Suite benoitbergeret.substack.com/…

"IA, emploi, modèles sociaux: le mirage américain" : ma réponse à @AndrewYNg open.substack.com/pub/benoit…

Jobpocalypse or Jobapalooza? The answer depends on the country. A (lengthy but hopefully useful) response to @AndrewYNg on artificial intelligence and work. open.substack.com/pub/benoit…

Growing synthetic humans open.substack.com/pub/benoit…

A revolution in synthetic data is coming. Not statistical samples of existing humans, but populations of synthetic humans grown from the behavioral rules anthropologists and behavioral economists have characterized for decades. Growing synthetic humans: open.substack.com/pub/benoit…

The air we will soon breathe open.substack.com/pub/benoit…

AI runs on data much narrower than most realise. The sky I read this morning, a mood across the table, the moisture in a vine, none of it lives in any system. Yet. Noticing piece, not analysis. New on Fault Lines 👇 The air we will soon breathe open.substack.com/pub/benoit…

our la première fois depuis Gutenberg, des cerveaux qui pensent en structures plutôt qu'en mots peuvent enfin écrire. Gaspard Koenig veut sanctuariser le livre contre l'IA. Il défend en réalité un monopole cognitif qui vacille. open.substack.com/pub/benoit…

February 28: joint US-Israeli strike, Khamenei killed. April 18: Palantir publishes a 22-point manifesto on X. 32 million views and still climbing. The text is drawn from Karp's 2025 book. Except the sharpest line in the tweet is not in the book. open.substack.com/pub/benoit…

Your school sold you a shell and called it a skeleton. Stanford's data shows what happens next. open.substack.com/pub/benoit…

IA générative : du nécessaire apprentissage de la mue pour naviguer l’ambiguïté du futur open.substack.com/pub/benoit…

OpenAI's new industrial policy: 13 pages on taxes and dividends. Zero on what its systems optimize for, filter, or erase. Everyone is debating the economics. The economics are the decoy. open.substack.com/pub/benoit… #openAI #policyproposal #ai

So, LLMs Have Emotions? We made this mistake with "intelligence." The word is the trap. New Fault Lines essay on why we need a different word, and what it might be. open.substack.com/pub/benoit…

LLMs treat silence as a bug, evasion as confusion, and ambiguity as noise. In half the world, those are core competencies. New piece on what the training pipeline can't see. open.substack.com/pub/benoit…

Un patient marocain qui répond "bessif" à un chatbot médical ne dit pas "oui, je suivrai le traitement." Le chatbot enregistre de la compliance. Le patient a dit tout autre chose. open.substack.com/pub/bberge… Sur ce que les modèles de langage font aux langues qu'ils ne voient pas.