Françoise Morvan

Via @visibrainFR #CES2025 - Zoom sur les influenceurs et journalistes francophones les plus engageants sur X 🔎

A 10-year-old child's experiment proved that memories can be passed down through generations. It all began when Jo Nagai, a second-grader from Kobe, Japan, noticed his hand-raised swallowtail butterflies seemed to recognize him, while wild ones flew away. Determined to find a scientific answer, he reached out to Georgetown University entomologist Dr. Martha Weiss, who had famously demonstrated that moths can retain memories through the chaotic cellular breakdown of metamorphosis. Inspired by her work, Jo proposed replicating her study on butterflies, leading to an extraordinary cross-continental collaboration. Using a home-designed setup, Jo trained caterpillars to associate a mild vibration with the scent of lavender. Astoundingly, after their brains were completely rebuilt in their chrysalis, 70 percent of the adult butterflies still avoided the lavender scent—proving their memories survived the transformation. But the most shocking discovery occurred when Jo bred these trained insects. Without ever experiencing the vibration themselves, both the offspring and the grandchildren of the trained butterflies inherited an instinctual aversion to the lavender scent. At just ten years old, Jo documented these revolutionary findings in a 33-page paper and presented them alongside Dr. Weiss at the International Congress of Entomology in Kobe. This extraordinary discovery of transgenerational memory not only challenges traditional understandings of genetics and inheritance but also highlights how the curiosity of a child can disrupt the boundaries of modern science. source: Upworthy. (2026). Ten-year-old butterfly researcher discovers that memories can be passed down through generations.

"Cela va changer fondamentalement la course à l'IA pour toujours": quelle est cette nouvelle intelligence artificielle originaire de Chine et qui impressionne les Américains? l.bfmtv.com/46QE

Sa collection de sacs à main se chiffre à 1,3 million de dollars: et si le footballeur norvégien Erling Haaland contribuait à sauver la maroquinerie de luxe? l.bfmtv.com/DX6b

👆🏽 Please see post 🎉 I am thrilled to share that I have received a 𝐃𝐢𝐬𝐭𝐢𝐧𝐠𝐮𝐢𝐬𝐡𝐞𝐝 𝐀𝐜𝐡𝐢𝐞𝐯𝐞𝐦𝐞𝐧𝐭 𝐀𝐰𝐚𝐫𝐝 in the 𝐓𝐡𝐢𝐧𝐤𝐞𝐫𝐬𝟑𝟔𝟎 𝐀𝐧𝐧𝐮𝐚𝐥 𝐀𝐰𝐚𝐫𝐝𝐬 𝟐𝟎𝟐𝟔 in two categories: 𝐈𝐧𝐟𝐥𝐮𝐞𝐧𝐜𝐞𝐫 𝐨𝐟 𝐭𝐡𝐞 𝐘𝐞𝐚𝐫 and 𝐀𝐮𝐭𝐡𝐨𝐫 𝐨𝐟 𝐭𝐡𝐞 𝐘𝐞𝐚𝐫! 🏆 I have also received an 𝐇𝐨𝐧𝐨𝐫𝐚𝐛𝐥𝐞 𝐌𝐞𝐧𝐭𝐢𝐨𝐧 in the 𝐓𝐡𝐨𝐮𝐠𝐡𝐭 𝐋𝐞𝐚𝐝𝐞𝐫 𝐨𝐟 𝐭𝐡𝐞 𝐘𝐞𝐚𝐫 category! 🏅 @terence_mills @thierry_pires @jeffkagan @JohnNosta @TylerCohenWood @tlloydjones @3BodyProblem @bulbi59 @FmFrancoise @Victoryabro @MaiaGabunia @AstridLavalette @amalmerzouk @postoff25 @vanivina92 @jeancayeux @renujaiho @NewsNeus @bbailey39 @smoothsale @ralf_ladner @ylecun @lindayaX @joana_ut @UNWomenWatch @X @DigitalEU #Thinkers360 #Awards2026 #ThoughtLeadership #ThoughtLeader #TechInfluencer #WomenInTech #B2BMarketing #InfluencerMarketing #DigitalTransformation #EmergingTech #Tech #TechCommunity

Many people are asking how did Kimi K3 catch up so fast to Western models? Simply, a frontier model really only requires compute, data and people. There is no magic secret. A few answers to this question 1. Kimi with help from the Chinese government has thousands if not tens of thousands of experts (lawyers, scientists, doctors, programmers.. etc.) making RL env data every day. A frontier model is RLed on “tasks”. Each one of these tasks needs to be created by either a human or an LLM. Claude did not wake up one day knowing how to use say the Github CLI. He learned how to use the Github CLI in an RL env.  Meta is pursuing this exact same strategy with its Applied AI org and IMO it appears to be working. 2. I have said this before regarding GLM 5.2 - Kimi obviously distilled from GPT 5.5 and Claude Opus. This only eliminated their cold start problem in RL, meaning they skipped say some X number of months in cold start RL. The mass number of RL envs created by their experts is still the most crucial part here, and you cannot attribute Kimi’s success to distillation. Distillation only saved them some time. 3. Agentic coding and frontier LLMs significantly accelerated their research. My hunch is they use proxies to access Claude API and GPT API for their own model development. Claude most likely wrote all their training code. This release leads to some very interesting questions. What happens now in a world in which an almost Fable class model is open sourced and free on July 27th? My view is intelligence / software will become very soon close to free. Chip makers and inference providers are big winners from Kimi’s success.

🤖 Pendant que l'Occident débat de la sécurité de l'IA, la Chine lâche des milliers de robots humanoïdes dans ses usines pour qu'ils apprennent « sur le tas ». Sa stratégie de formation : échouer vite, en public, à grande échelle. La donnée ne s'écrit pas dans un labo. Elle se ramasse par terre. 🔗 japantimes.co.jp/business/2026/… #robotique

Kimi K3 is changing how many people view Chinese AI and AI talents. And Kimi is not alone. GLM, DeepSeek, Qwen, and many other teams are showing that Chinese researchers are not simply adapting ideas from elsewhere. They are contributing directly to the progress of AI itself. People have shifted their question from “Can China build frontier models?” to “Who are the researchers pushing these ideas forward?” There is a small but revealing moment in Yang Zhilin’s talk. While explaining Kimi’s new Attention Residuals architecture, he pauses to mention Kaiming He’s ResNet tutorial at ICML in 2016. He remembers it because ResNet solved one of the biggest problems in deep learning: how to train much deeper neural networks. Before ResNet, adding more layers could make a network harder to optimize. ResNet introduced a simple shortcut that allowed information and gradients to flow more directly across layers. That simple idea became part of the foundation of modern AI, from AlphaGo to ChatGPT. Here is a great video about 大神 Kaiming He, the inventor of ResNet.

月之暗面新团队成员杨新宇解释了 Kimi 为什么能推出 K3。 他称，自己今年离开学界进入产业，期间接触了多家 AI 公司。他总结出同行的四种通病： ① 傲慢。 老牌团队认为 AI 战争已经结束，自己已经赢了，对未来和人才失去渴望。 ② 浮躁。 年轻实验室基础不牢，只顾追赶前沿，追不上就匆忙转向其他赛道。 ③ 胆怯。 一些团队经验丰富、实力不弱，却不敢把目标定为行业第一。 ④ 目标错位。 每个人都在争取个人功劳，却没人真正关心公司能否做出 AGI。 杨新宇称，月之暗面最不同的地方，是创始团队仍对 AGI 保持强烈追求。这也是他选择加入的原因。 他还晒出一张「Kimi 五戒」： ① 模型公司要做模型。 ② 做 Research、发 Paper 要做实验。 ③ 训练模型要看 Metric。 ④ 不 Work 的东西不要硬上。 ⑤ 不要 YOLO。 翻成大白话，就是少讲故事，多做实验。训练要看数据，失败方案及时停，也不要靠直觉豪赌。

Elon Musk just confirmed that SpaceXAI’s next 2-trillion-parameter model is already better than Grok 4.5 in every way and is expected to finish initial training next week Elon said it may surpass Kimi while keeping speed and token efficiency close to Grok 4.5 Grok 4.5 already delivers frontier intelligence with exceptional speed, remarkably low token usage and one of the lowest costs per task among leading models Its combination of intelligence, speed and efficiency is already in a league of its own Now SpaceXAI is scaling intelligence even further without abandoning the efficiency that made Grok 4.5 so impressive The next Grok model is not simply getting bigger It is getting smarter while staying fast and incredibly efficient

Never a more accurate depiction…

I stand by this: Kimi K3 will be a DeepSeek 2.0 moment in its own right. Perhaps not immediately visible on the stock market, but a global wake-up call that will trigger a "code red" for many.

Larry Ellison just described a world where intelligence is free. Ellison: “It will help us solve problems we couldn’t solve on our own… It’ll make us much better scientists, and engineers, and teachers, and chefs, and bricklayers, and surgeons.” Every profession. Every rung. All elevated at once. Sounds like progress. Until you realize what “elevating everyone equally” actually does. Humanity spent ten thousand years building everything on the assumption that knowledge is scarce, and mastery takes a lifetime to earn. AI doesn’t challenge that assumption. It erases it. When every surgeon operates with the same precision, twenty years of practice becomes invisible. When every engineer thinks at the same depth, experience stops compounding. When capability is universal, capability stops being what separates people. Ellison: “The smartest people I know are investing fortunes. Their fortunes. In building and training these AI models.” They’re not funding a tool. They’re funding the moment human performance gets a new floor. And they already see the question waiting on the other side: If everyone can think, what separates anyone? Not knowledge. That’s solved. Not skill. That’s compressed. What’s left is the one thing no model replicates. The problem you choose before anyone sees it. The question no dataset contains. The conviction to move while everyone else waits for proof. Ellison: “Is AI the most important technology in human history? We’re gonna find out soon. Well, it’s pretty clear.” It is clear. The largest leverage shift since language doesn’t destroy the edge. It moves it. From knowledge to vision. From skill to taste. From what you can do to what you decide is worth doing. This isn’t a threat to the ambitious. Not if they stop competing on what just became free. And start competing on what never will.

Kimi K3 may be super useful for finance, planning and operations, given how well it plays with spreadsheets. Took the top spot on SpreadsheetBench 2, almost tallying with Fable 5 and completing 34.8% of workflows. SpreadsheetBench 2 tests whether tool-using AI agents can finish entire spreadsheet jobs rather than isolated formulas. It measures autonomous workbook execution rather than general reasoning or everyday Excel assistance. It involves 321 expert-curated tasks spanning financial modeling, workbook debugging, and native chart creation. Each task averages 11.8 sheets and 593.5 cell changes, forcing long chains of dependent actions. Modeling and debugging tasks require every requested edit to match while untouched cells remain unchanged. Chart tasks pass only when a vision model confirms at least 70% of specified requirements.

L'Université UCLA 🇺🇸 a réussi à condenser un système complet d'ondes térahertz (connues pour leurs débits records et pouvant traverser des matériaux opaques), qui occupait en 1990 toute une table de laboratoire, sur une seule puce à semi-conducteur. samueli.ucla.edu/ucla-engineers…

Life with children

La startup 🇳🇱 Monumental conçoit des robots capables d’automatiser une partie de la construction de murs (pose des briques et du mortier). Sa flotte compte « 100 robots maçons », ayant déjà participé à la construction d'une centaine de maisons en Europe. usine-digitale.fr/intelligence-a…

Kimi K3 launched yesterday, and within hours our billboards across San Francisco had already been updated to reflect the new frontier The Artificial Analysis Intelligence Index moves as fast as AI does - whether you're checking our website or waiting for the 38 Geary.

Moonshot AI, the company behind Kimi, has four core founders. Their backgrounds are unusually strong: - Founder and CEO Yang Zhilin studied computer science at Tsinghua before earning his PhD from Carnegie Mellon. He was the first author of Transformer-XL and XLNet, and previously worked at FAIR and Google Brain. - Co-founder and CTO Zhang Yutao earned his PhD in computer science from Tsinghua. His earlier work covered knowledge graphs and AMiner, and he previously co-founded Recurrent AI with Yang. - Co-founder Wu Yuxin studied at Tsinghua and CMU before joining FAIR. He worked with Kaiming He on Group Normalization and also created Detectron2. - Co-founder Zhou Xinyu studied computer science at Tsinghua and later joined Megvii, where he worked on turning research algorithms into production systems and co-authored ShuffleNet. They all share one root: Tsinghua University. Tsinghua is widely regarded as one of China’s top universities. In the latest U.S. News Best Global Universities ranking, it reached No. 6 worldwide. Its influence on China’s AI industry extends well beyond Moonshot. Z.ai, the company behind the GLM models, also grew out of Tsinghua. Its co-founder and chief scientist, Tang Jie, was once Yang Zhilin’s teacher. There is also a more personal connection. Yang and Zhou formed a rock band together at Tsinghua. Moonshot AI’s Chinese name, 月之暗面, comes from Pink Floyd’s album The Dark Side of the Moon, one of Yang’s favorites. Kimi may look like a young AI company. Behind it is a much older network of classmates, teachers, research labs, and friendships.

A look into the Moonshot AI office, shortly before the Kimi K3 launch. History was being made!

Kimi's CEO Zhilin Yang: "everyone's trying to build one smarter agent many agent hits a wall fast, so instead of making it smarter we just made more - one boss, a thousand workers." in a 39-minute talk he explains why one agent won't get you to real work. agent swarms + long context + RL on every sub-agent - that's the fix. bookmark this ↓