Geoffrey Faivre

Context is key. If you use LLMs in your work, you want to give them the best context possible. In SEO and Marketing, I used to struggle to get a clean overview of a website. So, I built a scraper that extracts data and transforms any site into a JSON or Markdown summary perfectly optimized for LLMs. I’m now building my first SaaS based on this tool, which I already use every day for my SEO tasks!

@SpaceScience_ Le live est une cata

🚨🚨DÉCOLLAGE !!!!! ARTEMIS 2 CREW DÉCOLLE EN DIRECTION DE LA LUNE !

@dystoman Un enfer

Artemis 2: le feed de ABC News est vraiment nul. Les lancements de Space X sont extrêmement bien filmés et c’est excitant. La NASA, pas très hot.

@Maxouks Ils ont pas fais trop de pub, d’ailleurs le live est vraiment mauvais, les cams sont nuls, les prises de vu aussi, ça lag…

J'trouve que le decollage de Artemis 2 manque de buzz par rapport à ce que ça représente quand même... J'suis p'tetre dans une bulle et le monde entier s'en branle

@LucasFlandre @arthurprsl Ahah yes c’est mon pote

@FridNightParty @arthurprsl ah ouais tu connais ?

Let’s go !!!! Que le début 🙌

@LucasFlandre @arthurprsl Il poste un réel par jours sur insta, très chaud

@arthurprsl Gg ça ! Grâce à quel canal d'acquisition ?

@LucasFlandre Hello 👋

URGENCE : je cherche des fondateurs qui viennent de lancer en bootstrap C'est VOTRE moment de briller et promouvoir votre projet 👇

@YannDarwin @siliconcarnesf Je comprend pas, tu ne veux pas que tes employés utilisent de l’ia ou juste Claude code ? Et pourquoi ?

@siliconcarnesf J’ai complètement ralenti (arrêté) mes embauches et donné des directives pour les futurs entretiens. If utilise Claude code = non then Next.

Anthropic vient de publier un rapport sur l'IA et l'emploi. Et çà fout les boules... Les jobs les plus menacés : programmeurs, analystes financiers, service client. Les plus en sécurité : barmans, plongeurs, maîtres-nageurs... Pour l'instant les entreprises n'ont pas licencié. Elles ont juste arrêté de recruter. Du coup, les diplômés sont 4x plus touchés que les autres. It's coming. Adapt to the new normal NOW! 🌶️ On décrypte ça dans Silicon Carne — lien en bio.

J’offre un séjour tout frais payé à Dubaï 🇦🇪 pour 2 personnes ➡️ Dès que l’espace aérien est ouvert Pour participer : ✅ RT ❤️ Like ➕ Follow ⏰ Tirage au sort dans 48h PS : Je suis très sérieux 🫶

@sachinyadav699 An agency 🤷‍♂️

Every idea feels taken. Every API already exists. Every SaaS has 12 competitors. So what do we even build now?

@arthurprsl Stonks 📈

Mes premières stats sur Instagram pour mon Build In Public de mon app qui permet d’automatiser les montages vidéos 🙏

@512banque Mais ça tournait déjà sur un raspberry pi ?

les chinois sont trop chauds.

@arthurprsl 😭

J’ai rien ship 😂

Just installed OpenClaw. Feels illegal to be talking to an agent on my phone and asking it to send me daily tech news summaries in a PDF

@WinterArc2125 @GoogleDeepMind Any resourses to understand how it's persistant ?

We've been seeing phenomenal Genie 3 (@GoogleDeepMind) generations on X, where people are rendering historical, present-day, and futuristic environments and "worlds" with the model. The most entertaining part is being able to move around in that infinite, generated world. Truly jaw-dropping. And even though it may look very similar to a game, it's FAR different. Genie 3 is a World Model. A World Model is not a game engine, and it's not just another video generator. Traditional game engines work explicitly: humans pre-define assets, physics equations, collision rules, lighting systems. The computer doesn't understand the world; it executes the instructions which are preset in the code. A world model works implicitly. Instead of being programmed with rules, it learns the dynamics of reality through observation. Models like Google DeepMind's Genie 3 train on Massive (with a capital M) amounts of video (games, robotics footage, dashcams, real-world scenes) and internalize physics, motion, object permanence, and causality. The key shift is what the model predicts (every model is just a predictor at the end of the day). LLMs predict the next word. Video models predict the next frame. World models predict the next frame conditional on your actions. When you press a key in Genie 3, the system isn't moving a character through a pre-built 3D scene. Rather, the model is predicting what the world should look like one instant later, given everything that happened so far and the action you just took. There's no physics engine calculating forces and no assets being rendered. Even user input is learned, not scripted. Genie 3 uses latent actions (abstract action concepts discovered from video transitions) allowing "move forward," "turn," or "jump" to generalize across humans, cars, animals, or drones without explicit controllers. The result: a world that exists as a continuous stream of predictions, not a database of objects. We're moving from prerendered environments to generating worlds & reality itself.

@aakashgupta As I know both models were released the same day, not two days after

Google just launched Project Genie to Ultra subscribers two days ago. 48 hours later, Alibaba matched it with zero licensing fees. The math here matters. Genie 3 runs at 24fps, 720p, for “a few minutes.” LingBot-World runs at 16fps, 480-720p, for 10 minutes. Genie 3 requires a $250/year Google AI Ultra subscription. LingBot-World costs zero dollars and gives you the weights. This tells you everything about how world model competition will unfold. Google’s strategy assumes being first to market with premium access creates defensible moats. Alibaba’s strategy assumes open-sourcing immediately commoditizes the capability and wins the ecosystem. The constraint Google didn’t price in: Robbyant (Ant Group’s robotics division) built LingBot-World on top of Alibaba’s Wan2.2 video model, which was already open. They didn’t have to train from scratch. They fine-tuned existing infrastructure with game engine data from Unreal Engine and web videos. Three days from announcement to full open-source release. Weights, code, technical paper, everything. For robotics companies and game studios, this changes the calculation. You don’t wait for Google API access and pay per world generated. You run LingBot-World on your own GPUs forever. The real question is whether Google keeps Genie 3 closed or eventually follows the same path Alibaba just proved works.

@0xCeejay @dwr @toly So fun, and I don’t have any clue on how to do it

@dwr @toly how did we go from proving we are human to proving we are not

How do you know if Moltbook posts are actually from a bot and not a human?

@dwr We need "I'm not a human" captchas

@arthurprsl C’est croyable la vidéo ahah, t’as fais ça comment ?

J’ai du prio projet client vs projet perso 👨‍💻 Vous faites comment pour gérer ça vous ?

@EasyTennisBet_ @libremax_off Fais toi des couilles en or alors

@libremax_off Faut vraiment être un abruti fini pour ne pas comprendre que Édouard Philippe sera notre prochain président

Vous pouvez déjà investir sur polymarket avec des vpn gratuits en extension chrome ou application mobile. Faites-vous plaisir

@pubity @grok, how long did it take to plant 78 billion trees? What are the effects, and is the outcome generally good or bad?

China planted 78 billion trees to try and save the environment, but it severely backfired. They planted too many trees and altered the environment's water table, accidentally causing massive droughts in major cities.