🚀Robot capture des soldats russes, joue au ping-pong, plie le linge ; Is "mirror life" an existential risk? & more

Les 5 meilleurs cas d'usage de ChatGPT Agent Mode

Bonjour,

Vous recevez la version gratuite de la newsletter Parlons Futur : une fois par semaine au plus, une sélection de news, mêlant sources anglophones et francophones, résumées en bullet points sur des sujets tech 🤖, science 🔬, éco 💰, géopolitique 🌏 et défense ⚔️ pour mieux appréhender le futur 🔮.

Je m'appelle Thomas, co-fondateur de YeldaAI, qui développe des IA pour répondre au téléphone en langage humain pour les administrations et les entreprises. Plus d'infos sur moi en cliquant ici.

Mes derniers podcasts

• Sommes-nous prêts à nous entourer de millions de robots humanoïdes ? (Youtube, Spotify, Apple)

• L'IA une arnaque ? Pourquoi 42% des entreprises abandonnent leurs projets IA (YT, Apple, Spotify)

• AGI vs humains augmentés ? Qui va gagner ? (YT, Apple, Spotify)

• Faut-il interdire les IA pseudo-conscientes ? (YT, Apple, Spotify)

  • Résumé des arguments du directeur de Microsoft AI

• 🧠Contre-intuitif : pourquoi l'intelligence humaine baisse en fait depuis 10,000 ans (Youtube, Spotify, Apple)

• 🌍Entretien exclusif avec Philippe Bihouix, co-auteur de la BD à succès "Ressources : Un défi pour l'humanité" (Youtube, Spotify, Apple)

Et voici donc ma toute dernière sélection !

• In July, an Ukrainian Brigade carried out a historic fully autonomous operation, combining FPV drones and ground robots to force Russian troops to surrender in Kharkiv Oblast without any Ukrainian soldiers participating directly in the assault. (source)

  • Enemy fortifications were initially targeted by FPV drones and kamikaze ground robots. As another robot approached a destroyed bunker, a Russian soldier surrendered to avoid detonation. The surviving invaders were then guided to Ukrainian lines by drones and taken prisoner according to protocol.

• 4 new amazing new pics and videos from Starship's latest landing, here’s one:

  

• Scientists just developed a new AI modeled on the human brain — it's outperforming LLMs like ChatGPT at reasoning tasks (source)

  • Developed by Sapient, an AI company in Singapore, the hierarchical reasoning model (HRM) system is modeled on the way the human brain processes complex information, and it outperformed leading LLMs in a notoriously hard-to-beat benchmark.

  • The HRM model has 27 million parameters while using 1,000 training samples

  • In comparison, most advanced LLMs have billions or even trillions of parameters. GPT-5 is rumored to have 3 to 5 trillion parameters.

• Nvidia CEO Jensen Huang says AI will ‘probably’ bring 4-day work weeks (source)

  • ‘Every industrial revolution leads to some change in social behavior’

• What about these LLM that can ‘lie, steal and blackmail’? Tech analyst Ben Evans response:

  • "Anthropic made another play for regulatory attention with a paper claiming that LLMs will ‘lie, steal and blackmail’"

  • "It actually means that they asked an LLM what it would do in a situation where most people would do X, and the LLM returned the result ‘I would do X’. "

  • "Yes, well done. All of these stories have the same intellectual validity as writing ‘I’m alive!’ on a piece of paper, photocopying it, and then saying “Look! In our analysis, the photocopier says it’s alive!”"

• How GPT-5 came to be (source)

  • To train the new model, OpenAI used o3 to generate training data, as Sebastien Bubeck explained:

  • "We used OpenAI’s o3 to craft a high-quality synthetic curriculum to teach GPT-5 complex topics in a way that the raw web simply never could."

  • This could be repeated. Using the previous generation to create synthetic training data for the next. So far, there is no hard limit to how many times this can be done.

  • As long as each new teacher is stronger than the last and synthetic data is added rather than replacing the original, research suggests each generation of models should improve. Whether it can solve all remaining problems, like memory, hallucinations and context-management, remains up in the air.

• Google Deepmind CEO Demis Hassabis on their latest 3D Gen AI tool (source)

  • Simulations are the future, & one of the main tools we’ll ultimately use to understand and predict things about the universe. This is why I’m so excited about Genie 3, our latest interactive world simulator.

  • Genie 3 can accurately model intuitive physics - lighting, gravity, material, liquids… check out the realistic water movement when the character walks into the puddle! See

  • You can prompt Genie 3 with text, photos, or videos - like this cool game example created using Imagen 4 -> Veo 3 -> Genie 3: See

  • You can generate interactive 3D worlds with text, explore these worlds, even interact with objects in real time and navigate with keys: See

  • Genie 3 can bring awesome sci-fi artwork to life including details like lighting and shadows: See

  • Genie 3 has advanced spatial memory, when you make changes in the world, they persist in the simulation even when out of view! See

  • Control an avatar in real time (in this case a dog on a beach!): See

• Euh ??!! OpenAI, Google DeepMind and Anthropic sound alarm: 'We may be losing the ability to understand AI' (source)

  • This balance becomes more critical as AI systems become more powerful and potentially dangerous. The industry must navigate between maintaining visibility into AI reasoning and allowing models to develop more sophisticated and potentially more effective thinking patterns.

• Wild Video Shows Humanoid Robot Effortlessly Folding Laundry : watch the demo video, by AI robotics firm Figure

  • "first instance of a humanoid robot with multi-fingered hands folding laundry fully autonomously using an end-to-end neural network."

  • "Towels are deformable, constantly changing shape, bending unpredictably, and prone to wrinkling or tangling," argued Figure. "There's no fixed geometry to memorize, and no single 'correct' grasp point. Even a slight slip of a finger can cause the material to bunch or fall."

  • Figure also claims that its bot learned the skill without the company making any significant changes to its visual language system called Helix; on the contrary, it simply fed the model a new dataset.

• Un humanoide Unitree G1 entrainé à jouer au tennis de table : voir la démo de 33 secondes !

  • L’université de Berkeley dévoile HiTTER, un système qui permet à un humanoïde Unitree G1 de renvoyer une balle de ping-pong avec une précision et une réactivité inédites. (le robot est disponible depuis mai 2024)

  • à noter: des chercheurs aux US qui expérimentent avec des robots chinois... La Chine est en train de réussir son pari de s'imposer sur la hardware robotique

  • en tout cas bien plus impressionnant au ping-pong qu'au foot: regardez cette vidéo (gag) du tout dernier très sérieux tournoi de football de robots en Chine

    • This feels lighthearted. But under the hood is the race for a $15-40 billion market by 2030

• Le dernier modèle Unitree, version appelée R1, sorti en juillet 2025 : priced at under $6,000.

  • That’s not pocket change, but it’s orders of magnitude cheaper than most robots in its class, which can run into tens or even hundreds of thousands of dollars. (source, the 1-min demo)

  • The Unitree R1 is around 1.2 meters tall and weighs roughly 25 kilograms (similar to a packed suitcase). This makes it compact and relatively easy to move around.

  • It’s equipped with 24 to 26 degrees of freedom (think of these as “joints” that allow it to bend, twist, and rotate), giving it a surprisingly human-like range of motion. It can walk, squat, wave, balance, kick and—according to Unitree’s own demos—pull off athletic tricks like cartwheels.

  • but while the R1’s hardware is impressive, the software is still suboptimal, much of what we see in humanoid demos is either scripted routines or teleoperation (remote control). Mais son bas prix permet à d'autres chercheurs d'optimiser la partie logiciel, cf l'exemple ci-dessus !

• Et la version quadrupède d'Unitree, le Unitree A2 Stellar Hunter : vidéo impressionnante de 2 min

• Mode agent dans ChatGPT : les 5 meilleurs cas d'usage (JDN, avec prompts)

  • 1. La recherche d'emploi ou de candidats automatisée

  • 2. Veille concurrentielle et génération d'un rapport

  • 3. Automatisation de la recherche de donnée dans un tableur

  • 4. Identification des trends pertinentes sur les réseaux sociaux

  • 5. Un audit de conformité RGPD

• Bill Gerstenmaier, 71 years old, SpaceX's vice president of build and flight reliability

  • Il n'y a pas d'âge pour contribuer chez SpaceX !

  • Il y avait aussi eu Kairan Quazi, a young prodigy who was hired by SpaceX at the age of 14 in 2023, 55 ans d'écart entre ces deux employés dands cette boîte à la frontière de la tech...

• Google's Gemini reduced energy consumption per text query by 33x (!) in the last 12 month (source)

  • Currently averaging about 0.24 Wh of electricity per text query (equivalent to about 9 seconds of running a TV set)

• Open-source artificial intelligence models consume up to 10 times more computing resources than their closed-source competitors for the same task (source)

  • potentially undermining their cost advantages and reshaping how enterprises evaluate AI deployment strategies.

• New study finds AI could reduce global emissions annually by 3 to 5 billion tonnes of CO2-equivalent by 2035 (source)

  • Thanks to optimisations in transport, energy and manufacturing, offsetting way more than its own carbon footprint

• "Un pneu en 2020 contient deux fois plus d’ingrédients différents qu’une voiture entière un siècle plus tôt",

  • précise Jean-Baptiste Fressoz, historien des technologues et auteur du livre remarquable "Sans transition: Une nouvelle histoire de l'énergie"

• Une cyberattaque contre un hôpital britannique a contribué à la mort d’un patient (source)

• In America job postings for software developers have dropped by more than 66% since the beginning of 2022 (The Economist)

  • In January Marc Benioff, the boss of Salesforce, a maker of business software, said his firm will not add any more software engineers this year, owing to productivity gains from AI tools.

• Shadow AI economy booms while headlines about company-driven efforts cry failure (source)

  • A corporate lawyer quoted in the MIT report exemplified the pattern: Her organization invested $50,000 in a specialized AI contract analysis tool, yet she consistently used ChatGPT for drafting work because “the fundamental quality difference is noticeable. ChatGPT consistently produces better outputs, even though our vendor claims to use the same underlying technology.”

• AI Detects Hidden Lung Tumors Doctors Miss — And It’s Fast (source)

  • Et autre news troublante: Doctors Using AI Quickly Lose Ability to Spot Cancer, Study Finds (source)

    • In a new study, researchers found, upon surveying 19 doctors from four endoscopy practices between September 2021 and March 2022, that the clinicians seem to have gotten lazier and worse at their jobs as a result of the technology's assistance.

    • Researchers warned of a "deskilling" effect after noticing a 20 percent relative drop in the detection rate of adenomas, or pre-cancerous growths in the colon, after an AI was introduced at the centers at the end of 2021, compared to detection rates before the AI was implemented.

• Deepfake whales could be key conservation tool (source, relevée par Futur Proche)

  • Faute d’images réelles suffisantes pour entraîner les algorithmes, des chercheurs ont généré des clichés synthétiques de baleines noires de l’Atlantique Nord, une espèce en danger critique avec moins de 400 individus. « L’un des grands défis est le manque de données », explique un chercheur. Les modèles d’IA sont donc nourris avec ces deepfakes pour mieux repérer les cétacés sur images satellites ou drones et ainsi les surveiller.

• Cool: drone-powered sword that levitates towards you and flies around on command (15-sec video)

• Climate change could shut down ocean currents and freeze Europe (The Economist)

  • Brussels hitting -20°C in a bad winter. In Oslo the figure would be almost -50°C

  • Average rainfall in parts of northern Europe would drop precipitously; according to one estimate as much as 80% of England’s arable land would no longer be farmable without irrigation.

  • Storms would get worse; so, in some models, might summer heatwaves. This would be the worst of all worlds.

  • That would be very bad for the African countries on the south edge of the Sahara; it could also be devastating to the Amazon.

  • after a (currently unknown) temperature threshold is passed, the collapse could take just a few decades.

• Rock discovery contains ‘clearest sign’ yet of ancient life on Mars, NASA says (CNN)

Partagez cette newsletter par Whatsapp en cliquant ici ❤️

• Si vous appréciez cette première moitité de l’apéro gratuite, n’hésitez pas à prendre 3 secondes svp pour transférer cet email ou l’envoyer à ne serait-ce qu’un contact par Whatsapp en cliquant ici 🙂

• Et si on vous a partagé cet email, vous pouvez cliquer ici pour vous inscrire et ne pas manquer les prochains

Gemini Nano Banana : comment utiliser l'outil de retouche photo de Google ? (JDN)

• Sorti le 26 août dernier, Gemini 2.5 Flash Image permet de créer des images de haute qualité dans un large éventail de styles.

• Mais ses capacités les plus impressionnantes restent l'édition et la modification de visuels déjà existants.

• directement disponible dans l'application Gemini ou depuis Google AI Studio.

• permet notamment de :

  • fusionner plusieurs images en une seule

    

    Le résultat est visuellement très correct et aurait pris plusieurs minutes avec outil d'édition de type Photoshop. Le plus impressionnant étant que l'ombre de la voiture est parfaitement cohérente.

  • générer des variations dans une image en maintenant la cohérence des objets ou personnes, comme changer l'angle d'une photo !

  • transformations ciblées (floutage, découpage, suppression d'un objet / d'un humain, modification des paramètres de luminosité et colorimétrie).

• How to best prompt it: Google's tutorial

• A user created a 30-sec manga video in 2 hours using Google's text-to-image Nano Banano AI and Bytedance's new flagship image‑to‑video software Seedance 1.0

  • See the video and the step-by-step tutorial

---

👋👀🦉💡 Je vous invite maintenant à vous abonner à la version payante (l’équivalent du prix d’un café par semaine, le minium imposé par Substack) pour recevoir la semaine prochaine les résumés de ces longs articles issus des plus grands journaux (et accéder aux futures éditions payantes et archives) en cliquant sur le bouton “Upgrade to paid/Subscribe now” 👇

• 🔒 L'incroyable future mission robotique Dragonfly qui partira pour Titan, lune de Saturne, en 2028

• 🔒 Why The Fever Dream of Imminent Superintelligence Is Finally Breaking

• 🔒 How The Age of Robotic Warfare Finally Arrived

• 🔒 AI’s Overlooked $97 Billion Contribution to the Economy

• 🔒 Why a famous actor sent to space by Jeff Bezos said ‘When I looked … into space, there was no mystery, no majestic awe to behold … all I saw was death.’

• 🔒 Why Microsoft AI director Mustafa Suleyman says we should ban seemingly conscious AI

• 🔒Why Scientist Warns That New Synthetic "Mirror life" Could Spell Doom for Humankind "

• 🔒 How the Discovery of a Secret Superpower in the Bee Brain Could Revolutionize AI and Robotics

Les dernières newsletters :

• 10 crazy Veo 3 AI videos ; lentilles pour voir dans le noir ; OpenAI co-founder: "we need a bunker before AGI"

• AI existential risk; robot mille-pattes; moto volante; robot devient fou; 1-gram insect robot; AI tries to blackmail engineer & more

• Best evidence yet of alien life, prehistoric wolf resurrected, the 4 ways AGI may destroy us, talking to dolphins & more

• The best voice AI? ; why Tesla may still win ; AI keeps cheating ; one human controls 100 robots ; crazy moon landing video & more

• Les précédentes

Merci, et à bientôt !

Thomas