Kent Beck on "features vs futures", and why AI makes it more urgent, not less. Every feature you ship spends some of your futures: the options you still have for what the software could do next. Keep spending and you end up with a codebase where nothing can change without breaking something else. It used to take a hundred people ten years to get there. With AI, he says, one person can do it in a week. His answer is to rest between the notes. When the genie says "done, want the next feature?", that's the moment to pause, refactor, remove duplication, even throw it away and try again. The value of a project is its features plus its futures, and you can add value by adding futures. A ~5 min excerpt from his Prodacity 2026 talk (Rise8). Summary is my paraphrase, so check it against the clip. Full talk: #softwareengineering #ai #refactoring

At the Valdai Club, Rick Sanchez put a pointed question to Vladimir Putin: Russia was ready to deal at Minsk, at Istanbul and at Anchorage and was turned down each time. Now Russia is gaining ground. So why negotiate at all? Putin's answer: it isn't about the piece of land. Around 11% of the Donetsk People's Republic is still held by Ukraine; Kyiv says it can hold it for a year and a half, Europe says two, and Moscow expects it much sooner. If that territory is going to be lost anyway, why keep sacrificing lives for it? What Russia wants, he says, is not a ceasefire but a durable peace: security guarantees for both sides, protection for the Russian language and the Russian Orthodox Church in Ukraine, and an end to efforts to separate the two peoples for good. Russia is ready to start negotiations and finish them quickly, on terms acceptable to the Russian people. 5-minute excerpt, via the English interpretation. Full exchange: #russia #ukraine #putin

Des modèles qui passent l'examen du barreau, écrivent des dissertations et du code. Et pourtant, demande Yann LeCun à Sciences Po, où est ma voiture autonome ? Où est le robot qui débarrasse la table ? Pour lui, les LLM entraînés sur du texte ne sont pas un chemin viable vers une intelligence de niveau humain. Le texte, c'est la partie facile ; le monde physique est bien plus compliqué, et aucun système actuel ne le comprend aussi bien qu'un chat de gouttière. Son calcul : tout le texte public d'internet, environ 10^14 octets, c'est à peu près ce qu'un enfant de 4 ans a reçu par le seul nerf optique en 16 000 heures d'éveil. Une grande partie de la réalité n'est tout simplement pas dans le texte. Extrait de 4 minutes, résumé à vérifier contre la vidéo. Conférence complète : https://www.youtube.com/live/Y4s8NadbZfU #ia #llm

Many people want to destroy Russia more than they want to save Ukraine: that was Julija Mendel's message to MEPs at the European Parliament. She was Zelensky's press secretary for two years. Her argument: a third of Ukraine has become hell, nearly 20% occupied and another 10% a burning front, and that daily suffering is weighed against a grand idea, Russia's collapse, whose price is her country. Then the numbers. Echoing JD Vance's essay that the maths in Ukraine doesn't add up, she asks how 25 million Ukrainians, half of them pensioners, are meant to defend 500 million Europeans against 140 million Russians. By her figures the population has fallen from over 52 million to around 25 million, with deaths running three to four times births. Simple maths, she says, showing a country disappearing. The figures are hers. A 3-minute excerpt from a 36-minute speech, German voice-over (my summary, check it against the clip). Full speech: #ukraine #peace #europe

What is the "foot-to-glute connection"? Coach Chong Xie explains it to John Stockton on The Ultimate Assist. His account: the fascia under your foot is full of sensors, and heavily cushioned shoes leave them under-stimulated. His test is simple: stand barefoot on a towel with your heel slightly raised and grip and release with your toes. He says elite athletes feel it travel up to their glutes within seconds, while most people only feel it in the calf, hamstring, shin or the foot itself. His conclusion is that top athletes got there by developing that link from the foot up, not by training glutes, squats and deadlifts in isolation. These are his claims, not settled science. A 3-minute excerpt (my summary, check it against the clip). Full interview: #fascia #athletics #training

Why does burnout feel so sudden? Anaesthetist Dr. Tim Hughes explains it with Christine Miserandino's "spoon theory": each day you get a fixed number of spoons, units of energy, and everything you do costs some. His example: a theatre list that used to take five spoons, with a familiar surgeon and team, now takes fifteen or twenty once staff are off sick and you're working with people you've never met. His answer is to cut the cost with slow nasal breathing through the day, and to win spoons back with a short guided breathing session at lunch. He says his wearable now shows less cardiovascular strain on on-call days that look exactly the same. 3-minute excerpt from his conversation with Patrick McKeown (my summary, check it against the clip). Full episode: #breathing #burnout #medicine

Faut-il avoir élevé un enfant pour briguer l'Élysée ? Dans sa chronique sur Tocsin, Thomas Séraphine propose d'en faire une condition pour se présenter à la présidentielle, qu'il s'agisse d'un enfant biologique, adopté ou de celui d'un conjoint. Son argument : un enfant vous rend vulnérable. Quand on a des enfants, la guerre cesse d'être des flèches rouges sur une carte de plateau télé. Il l'illustre avec Kipling, qui fit engager son fils John, tué en 1915, et David Grossman, qui appelait au cessez-le-feu quelques jours avant que son fils Uri ne tombe au Liban. Sa conclusion : ceux qui dirigent devraient savoir ce qu'il en coûte de voir partir ses enfants. Extrait de 2 min 44 d'une chronique de 13 minutes (résumé de ma part, à vérifier dans la vidéo). Chronique complète : #nostrfr

Pierre Conesa, ancien haut fonctionnaire du ministère de la Défense, sur la fabrication de l'ennemi. Son point de départ est cartographique. La Terre est ronde, les cartes sont plates, et une carte nationale française place la France au centre des terres habitées. La représentation géopolitique, chez les décideurs comme dans l'opinion, se construit là-dessus. Vient ensuite le discours. Pour faire l'unité nationale, il faut une cause commune, et cette cause prend souvent la forme d'une menace. Il rappelle au passage que le terrorisme est un mode d'action, pas un ennemi. Ce qui rend l'argument utile, c'est qu'il fonctionne dans les deux sens. Les Russes ont un complexe d'encerclement, chacun voit d'abord ses propres menaces. Mais il y a aussi, dit-il, quelqu'un d'assez imprévisible à Washington : envahir le Venezuela pour enlever un chef d'État, venant de Poutine, on aurait crié au scandale. Sa conclusion est constructive plutôt que résignée. On sait fabriquer un ennemi, on sait aussi défaire la chose : il a fallu trois guerres franco-allemandes pour arriver au traité de l'Élysée. Les Européens ont intérêt à développer une pensée stratégique propre, avec cette difficulté qu'un État balte et l'Espagne ne regardent pas la même carte. Extrait de trois minutes d'une émission de 46 minutes sur Tocsin. Émission complète : #nostrfr #geopolitique #russie

A new model called Jev can't write. Not "writes badly" — it cannot return a sentence at all. You hand it some text and a list of the answers it's allowed to give, and it picks one. That is the entire product. Nate B Jones's point is what happened next. TypeSafe shipped it, and he says that within 24 hours it was the fastest-adopted model in Vercel's AI gateway history — faster than any LLM from any lab, with more than twice as many paying teams as any previous launch managed in its first day. His conclusion: either a lot of good developers lost their minds simultaneously, or the thing Jev refuses to do is exactly the point. A great deal of what an AI pipeline actually does is classify, route and decide — work that never needed prose, and that you have been paying an LLM to wrap in sentences anyway. One-minute excerpt from a 33-minute video by Nate B Jones. Full version: #ai #llm #jev

Until 2014, nobody had run an empirical test on how bank lending actually works. Richard Werner ran it. The claim he was testing — that a bank creates new money when it lends, rather than passing on somebody else's deposit — was treated as a conspiracy theory at the time. Textbooks didn't entertain it, and economists called the people making it cranks. The textbook account is still the financial intermediation theory: banks gather deposits and lend those same deposits out. Werner's argument is that this is simply not what happens. His explanation for how it stayed hidden is the part worth the two minutes: double-entry bookkeeping. Every account balances, so there is never a visible moment where money appears — look at the ledger and the honest question is “show me where the money creation is”. He argues that worked well enough that mainstream economists went on believing banks don't create money right into the 21st century. Excerpt from a much longer conversation with Peter McCormack. Full interview: #banking #economics

À Sciences Po, on demande à Yann LeCun ce qu'il pense de la pause dans le développement de l'IA. Il ne prend pas de pincettes. Son raisonnement, en bref : ceux qui réclament une pause ne sont pas cohérents. Il remonte à 2019, quand OpenAI jugeait GPT-2 trop dangereux pour être diffusé — avant de le publier six mois plus tard, puis de garder GPT-3 fermé pour des raisons qui tenaient davantage de l'avantage commercial que de la sécurité. Le plus intéressant, c'est là où il arrive : l'argument selon lequel l'IA serait trop dangereuse pour être mise entre toutes les mains, et qu'il faudrait donc la réguler et interdire l'open source — sauf pour ceux qui la développent déjà — relève moins, selon lui, de la sécurité que d'une tentative de capture du marché par la régulation. Extrait de 3 min d'une conversation de 90 minutes à Sciences Po. Une digression sur le financement d'Anthropic a été retirée au montage. Conférence complète : https://www.youtube.com/live/Y4s8NadbZfU