Sabitlenmiş Tweet
Stew
2.9K posts

Stew retweetledi
Stew retweetledi

Lex Fridman podcast, every single episode:
Lex: My guest today invented modern computing. Before we start. What is a computer?
Guest: Well it's a machine that
Lex: But what IS a computer. Is a rock a computer.
Guest: No.
Lex: The atoms inside the rock are computing.
Guest: That's not how any of this
Lex: Do you love your work.
Guest: Sure?
Lex: On a scale of 1 to 10 how much do you think Stalin loved his work.
Guest: What
Lex: We're 4 hours in. I want to ask you about consciousness.
Guest: You haven't let me finish a sentence yet
Lex: Beautiful. Beautiful question my friend.

English
Stew retweetledi
Stew retweetledi

Nine billion miles of driving data just became a chip.
Tesla AI5 is finalized for production. The design files are at Samsung in Texas and TSMC in Arizona. The transistors are locked. There is no going back. Tape-out is the hardest gate in semiconductor engineering because everything before it is reversible and everything after it is silicon.
How this particular chip was designed is the most interesting part.
Nvidia builds general-purpose GPUs. They pack transistors into a full-reticle die, ship it with CUDA, and let customers figure out which operations matter. Blackwell B200 delivers 4.5 petaFLOPS at up to 1,000 watts. It runs any model for any customer. That generality is the moat and the tax. Every workload pays for circuits it never uses.
Tesla designed AI5 backward. They started with 9 billion miles of FSD inference data and asked one question: where does the neural network waste cycles? The answer was softmax computation and quantization precision loss. Two specific mathematical operations that consume disproportionate silicon area and power in every general-purpose GPU on Earth. Operations that Nvidia cannot optimize away because other customers need those transistors for different workloads.
Tesla hardened them. Burned custom quantization and softmax accelerator blocks directly into the die. Five times more efficient on those operations than any general-purpose equivalent. Then they added 10 times the raw compute and 9 times the memory capacity relative to AI4. The result: a single AI5 system-on-chip delivers roughly 5 times the useful compute of the current dual-chip AI4 configuration at an estimated 250 watts. Musk has framed a single AI5 as Nvidia Hopper class and dual AI5 as Blackwell class for Tesla workloads, at 3 to 5 times better power efficiency and roughly 10 times better performance per dollar.
This is not a chip designed to compete with Nvidia. This is a chip designed to run one thing: the learned differentiable physics engine that emerged from 9 billion miles of camera observation. Every transistor serves that engine. No wasted silicon. No generality tax. The neural network wrote its own hardware.
The chip goes to two foundries. Samsung in Taylor, Texas. TSMC in Arizona. Both American. Musk thanked both this morning and added: “It will be one of most produced AI chips ever.”
Samples arrive late 2026. Volume targets H2 2027. In the same post, Musk confirmed AI6, Dojo3, and “other exciting chips” are in active development. The 9-month cadence is real. AI6 targets tape-out by December. Dojo3 restarts on the unified architecture after Musk shut down Dojo2 last August as an “evolutionary dead end.” Intel joined Terafab eight days ago for advanced packaging. The $16.5 billion Samsung deal runs through 2033.
The chip that taped out this morning is not a product. It is the physical crystallization of 9 billion miles of learned physics into transistors optimized for the exact mathematical operations that physics requires. The software trained on the road. The silicon was designed from what the software learned. And the factory that will mass-produce it is being built in the same city where the cars that generated the training data roll off the line.
The loop is closed.

English
Stew retweetledi

Je crois qu'on ne mesure pas ce qu'Elon Musk est en train de construire avec X.
Tous les médias de l'histoire ont été couplés à une culture, une langue, une bulle géographique. Le Monde parle aux Français. Le NYT parle aux Américains. NHK parle aux Japonais. Chaque média filtre le réel à travers le prisme de sa culture locale.
X est en train de devenir le premier média de l'humanité. Pas d'un pays. De l'espèce.
Je le vis en temps réel. Mes posts en français se font RT par des Japonais, répondre par des Brésiliens, citer par des Américains. Des conversations qui n'auraient jamais existé il y a 5 ans. Un libertarien français qui débat avec un ingénieur de Tokyo et un entrepreneur de Sao Paulo sous le même tweet. Pas traduit par un éditeur. Traduit instantanément par l'IA, en un clic.
Les bulles de filtre culturelles sont en train d'exploser.
Et je pense qu'on sous-estime massivement les effets composés de ça.
Quand une idée peut traverser un océan en 3 secondes, quand un argument sourcé posté à Paris peut être vérifié par un économiste à Singapour et amplifié par un développeur à Austin dans la même heure, le coût de propagation d'une bonne idée tend vers zéro.
Et c'est catastrophique pour un type d'acteur très précis : les médias qui ont construit leur business model sur le monopole de l'information locale. Ceux qui pouvaient raconter n'importe quoi sur "ce qui se passe ailleurs" parce que personne ne pouvait vérifier.
Quand un journaliste français écrit que "le modèle américain ne marche pas", maintenant il y a 50 Américains dans les réponses avec des sources. Quand un éditorialiste dit que "le Danemark prouve que le socialisme fonctionne", il y a un Danois qui explique que le Danemark est 10e en liberté économique mondiale.
Le fact-checking n'est plus un département. C'est un effet réseau.
Les médias honnêtes n'ont rien à craindre de ça. Les médias qui vendaient une narration protégée par l'ignorance géographique de leur audience vont avoir un problème existentiel.
Parce qu'on ne peut plus mentir à l'échelle locale quand le monde entier regarde.
Français
Stew retweetledi

hey ramon, use this shit on hn and make me a newspaper

Cloudflare Developers@CloudflareDev
Introducing the new /crawl endpoint - one API call and an entire site crawled. No scripts. No browser management. Just the content in HTML, Markdown, or JSON.
English

Apple should make a Macbook in British racing green.


GregsGadgets@GregoryMcFadden
Hey Apple, see how everyone loves the colors on the Neo and how the iPhone 17 Pro is selling really well? I have an idea for you, free of charge.
English
Stew retweetledi
Stew retweetledi















