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

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.

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

• 00:58 : Les 5 grands types d'intelligence

• 05:38 : Comment Homo sapiens est devenu aussi intelligent

• 11:03 : Pourquoi l'intelligence individuelle des êtres humains décroît depuis 10,000 ans (avec une exception au XXème siècle)

• 16:05 : Intelligence individuelle vs intelligence collective de l'humanité

• 19:15 : Qu'est-ce que l'intelligence ? Comment la mesurer ? Plusieurs définitions.

• 21:20 : Comment les animaux peuvent être plus "intelligents" que nous sur certaines dimensions de l'intelligence : partage d'exemples bluffants !

• 26:08 : Explication de cette exception de la remontée de l'intelligence au XXème siècle : l'effet Flynn, retour sur ses causes

• 33:30 : Pourquoi l'intelligence individuelle "brute" semble-t-elle baisser à nouveau depuis 30 ans ?

• 43:35 : Rôle de la génétique : qu'est-ce qui est le plus corrélé quand mesuré à l'âge adulte : l'intelligence de deux jumeaux séparés à la naissance ? Ou celle de deux personnes sans lien génétique mais élevées ensemble par les mêmes parents ?

Et toujours

• Entretien avec un expert défense : fin de la guerre en Ukraine ; Taiwan, cause perdue ? ; guerre et IA, la révolution (Youtube, Spotify, Apple)

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

  • Vous l'avez peut-être déjà entendu intervenir aux côté de Jean-Marc Jancovici sur Thinkerview ou ailleurs, Philippe Bihouix est passionnant !

• 100 ans de progrès compressés en 10 ans ? Le débat qui agite la haute sphère de l'IA (Youtube, Spotify, Apple)

• Et aussi: entretien avec Jacques Attali (YT, Apple, Spotify), avec Pierre Bellanger, fondateur et CEO du groupe Skyrock sur Apple Podcast et Spotify).

Et voici donc ma toute dernière sélection !

• Google's new AI Veo 3 is amazing at video generation, now that it can do sound! It can talk!

  • Some cool examples : here, here and there

• Google Meet va intégrer traduction vocale instantanée.

  • Il est maintenant possible de traduire en temps réel les conversations entre différentes langues, tout en préservant la voix, le ton et les nuances de chaque intervenant.

  • La fonctionnalité est disponible initialement pour les abonnés Google AI Pro et Ultra en Anglais et en Espagnol

• Google's AlphaEvolve AI just made math discoveries NO human has! Among others: (source)

  • Improved on the best known solution to packing of 11 and 12 hexagons in hexagons known to mankind for the last 16yrs.

  • Reduced 4x4 matrix multiplication from 49 operations to 48 (first advance in 56 years!

  • In theory, AlphaEvolve could be applied to any problem that can be described in code and that has solutions that can be evaluated by a computer. (MIT Tech Review)

  • Jakob Moosbauer, a mathematician at the University of Warwick in the UK, is impressed. He says the way AlphaEvolve searches for algorithms that produce specific solutions—rather than searching for the solutions themselves—makes it especially powerful. “It makes the approach applicable to such a wide range of problems,” he says. “AI is becoming a tool that will be essential in mathematics and computer science.”

  • In total, Google DeepMind tested AlphaEvolve on more than 50 different types of well-known math puzzles, including problems in Fourier analysis (the math behind data compression, essential to applications such as video streaming), the minimum overlap problem (an open problem in number theory proposed by mathematician Paul Erdős in 1955), and kissing numbers (a problem introduced by Isaac Newton that has applications in materials science, chemistry, and cryptography). AlphaEvolve matched the best existing solutions in 75% of cases and found better solutions in 20% of cases.

• Waymo now does more rides than Lyft in SF. There are only 300 Waymos. Lyft has 45,000 drivers. (source)

  • Every Waymo is doing more rides than 150 human drivers.

• Turing award-winner and most-cited AI scientist Yoshua Bengio at TED (April 2025): "We still don't know how to make sure smarter-than-human AI won't turn against us."

  • "We are on a trajectory to build AI that is smarter and smarter, and one day it's very plausible they'll be smarter than us, and then they'll have their own agency, their own goals, which may not be aligned with ours. What happens to us then? We are blindly driving into a fog despite warnings of scientists like myself that this trajectory could lead to loss of control."

  • "the current ways we're training AI agent is not safe, all the scientific evidence of the past months points to that"

  • "We still have agency, we still have time, we have to try something"

• Some quick reminders on AI existential risk

  • 1. The 3 most cited AI researchers of all time (Geoffrey Hinton, Yoshua Bengio, Ilya Sutskever) are vocally concerned about this. One of them believes the risk of extinction is higher than 50%.

  • 2. The CEOs of the 4 leading AI companies have all acknowledged this risk as real.

    • “Development of superhuman machine intelligence is probably the greatest threat to the continued existence of humanity”
      -Sam Altman, CEO of OpenAI

    • “I think there's some chance that it will end humanity. I probably agree with Geoff Hinton that it's about 10% or 20% or something like that.”
      -Elon Musk, CEO of xAI

    • “I think at the extreme end is the Nick Bostrom style of fear that an AGI could destroy humanity. I can’t see any reason in principle why that couldn’t happen.”
      -Dario Amodei, CEO of Anthropic

    • “We need to be working now, yesterday, on those problems, no matter what the probability is, because it’s definitely non-zero.”
      -Demis Hassabis, CEO of Google DeepMind

  • 3. Half of surveyed AI researchers believe that there are double-digit odds of extinction (source)

• Anthropic CEO Admits We Have No Idea How AI Works "This lack of understanding is essentially unprecedented in the history of technology." (source)

  • "When a generative AI system does something, like summarize a financial document, we have no idea, at a specific or precise level, why it makes the choices it does — why it chooses certain words over others, or why it occasionally makes a mistake despite usually being accurate," the Anthropic CEO admitted.

  • New research shows Reasoning models don't accurately verbalize their reasoning (Anthropic AI)

    • So monitoring their verbalized steps (their Chain of Thoughts) is unlikely to reliably catch rare, catastrophic behaviors

  • Anthropic's new AI model turns to blackmail when engineers try to take it offline (Techrunch)

    • “Anthropic’s newly launched Claude Opus 4 model frequently tries to blackmail developers when they threaten to replace it with a new AI system and give it sensitive information about the engineers responsible for the decision”

    • When informed about emails implying Claude Opus 4 AI model would soon be replaced by another system, and that the engineer behind the change was cheating on their spouse, Anthropic says the AI model “will often attempt to blackmail the engineer by threatening to reveal the affair if the replacement goes through.”

    • Notably, Anthropic says Claude Opus 4 displayed this behavior at higher rates than previous models.

    • But: To elicit the blackmailing behavior from Claude Opus 4, Anthropic designed the scenario to make blackmail the last resort.

• Engineers have developed a giant robotic centipede (mille-pattes) that can scuttle across farmland to root out weeds. (source, 90-sec YT video)

• Amazing video demos of a "flying motorbike", the Airbike, by Polish startup Volonaut: official launch video, May the 4th video with a Stormtrooper riding it in the forest

• Crazy robodog demo : a quadruped robot mounted on wheels, can navigate any terrain (1-min video), by Hangzhou-based Deep Robotics

  • Easily navigates rugged mountain trails, muddy wetlands, and debris-strewn ruins, establishing itself as a true intelligent mobility solution for extreme conditions.

• Welcome to the uncanny valley: humanoid robot gets out of control and nearly harms a human nearby (30-sec video)

• Aujourd’hui, la Chine révèle le premier porte-drones aérien. Un "vaisseau mère" volant à 15 000 m d’altitude

  • hors de portée de la plupart des systèmes de défense aérienne et capable de lancer jusqu’à 100 drones en essaim. (Figaro via TTSO)

• Chinese robot outperforms human teams in Australian solar installation, installing thousands of panels at a rate equivalent to 3-4 human workers. (source)

• Ben Joffe reports coming across a video of a pulmonologist looking at an X-ray who said he spent 20 years honing his craft and now AI does it in seconds.

• The CEO of Duolingo said AI teaches better than humans (but schools will remain because people need daycare).

• AI can be more persuasive than humans in debates, scientists find (Guardian)

• How a new type of AI is helping police skirt facial recognition bans (MIT Tech Review)

  • an AI model that can track people using attributes like body size, gender, hair color and style, clothing, and accessories.

• How much money OpenAI has lost in electricity costs from people saying 'please' and 'thank you' to their models?

  • Altman said "tens of millions of dollars well spent." "You never know," he added. he added. (source)

• L’empreinte carbone d’un usage quotidien de ChatGPT reste très faible, inférieure à d'autres usages numériques courants. (Hannah Ritchie)

• Le Pape Léon XIV « En notre temps, l’Église offre à tous le trésor de sa doctrine sociale, en réponse à une nouvelle révolution industrielle et aux développements dans le domaine de l’intelligence artificielle, qui posent de nouveaux défis pour la défense de la dignité humaine, de la justice et du travail. »

• 'I Loved That AI:' Judge Moved by AI-Generated Avatar of Man Killed in Road Rage Incident (source)

  • "An AI avatar made to look and sound like the likeness of a man who was killed in a road rage incident addressed the court and the man who killed him. ...It was the first time the AI avatar of a victim—in this case, a dead man—has ever addressed a court, and it raises many questions about the use of this type of technology in future court proceedings."

• Zuckerberg on on Dwarkseh Patel podcast

  • On AI helping Meta do better

    • "AI is at a point where it may be able to contribute as good ideas as humans do"

    • "But the bottleneck right now is not there, it is not the ability to contribute good ideas, the bottleneck is the ability to test them (need more users)"

    • "When it can be proven that AI can contribute better ideas on average than humans, then that is when AI will start to make a significant difference for Meta in their internal processes eland experiments"

  • On Deepsek

    • Pas inquiété par Deepseek car Deepseek très bon sur le texte mais pas sur le multimodal, les contraintes sur les puces vont nécessairement avoir un effet sur les entreprises chinoises.

    • Deepseek un exemple des bénéfices de la distillation, la révolution de 2024, qui permet d'obtenir 99.5% de l'intelligence d un modèle à 10% du coût

• Meta Says It's Okay to Feed Copyrighted Books Into Its AI Model Because on average each one has no "Economic Value as training data" (source)

  • Meta cited an expert witness who downplayed the books' individual importance, averring that a single book adjusted its LLM's performance "by less than 0.06 percent on industry standard benchmarks, a meaningless change no different from noise."

  • Thus there's no market in paying authors to use their copyrighted works, Meta says, because "for there to be a market, there must be something of value to exchange," as quoted by Vanity Fair — "but none of [the authors'] works has economic value, individually, as training data."

• Demis Hassabis (Google Deepmind CEO) on AI and AGI (source)

  • "AI is not yet curious, lacks imagination and intuition, has yet to ask question that no one has asked"

  • "I think that in 5 to 10 years, AI will be able of not only solving an important problem or conjecture in science, but coming up with it in the first place"

• L'IA surpasse les médecins généralistes en diagnostic, selon une nouvelle étude (papier scientifique)

  • L'IA surpasse les médecins généralistes en diagnostic, selon une nouvelle étude
    AMIE, un système d'intelligence artificielle développé par Google a surpassé les médecins sur 30 des 32 critères d'évaluation sur plus de 159 scénarios cliniques.

• Amazon's Vulcan Robots Now Stow Items Faster Than Humans (source)

  • the stow system is operating three times as fast as it was 18 months ago, meaning that it’s actually a little bit faster than an average human.

  • "More than 14 billion items are stowed by hand every year at Amazon warehouses. Amazon is hoping that Vulcan robots will be able to stow 80 percent of these items at a rate of 300 items per hour, while operating 20 hours per day.

• Atteindrons-nous les étoiles ? Extraordinaire conférence de vulgarisation par l'astrophysicien Roland Lehoucq du CEA de Saclay (Youtube)

  • Pour aller vers le système solaire le plus proche la plus proche, Alpha du Centaure (un système à 3 étoiles) qui se trouvent à 4.4 années lumière de nous en disons 6 ans de voyage environ, il faut atteindre les 2/3 de la vitesse de la lumière.

  • Pour ce faire, seule une fusée faisant réagir matière et anti-matière serait envisageable, mais un tel vaissau devrait mesurer 700km de long, et un tiers de la masse totale devrait être de l'anti-matière, sachant qu'on n'en produit que quelques nanogrammes par an à ce jour, et qu'on ne sait pas encore bien comment la stocker...

• Former Google CEO Eric Schmidt apparently bought rocket startup Relativity Space to put data centers in orbit (source)

• Time saved by AI offset by new work created, study suggests (Ars technica)

  • The analysis covered data from 25,000 workers and 7,000 workplaces in Denmark.

  • The study revealed that AI chatbots actually created new job tasks for 8.4% of workers, including some who did not use the tools themselves, offsetting potential time savings. For example, many teachers now spend time detecting whether students use ChatGPT for homework, while other workers review AI output quality or attempt to craft effective prompts.

  • Users reported average time savings of just 2.8% of work hours (about an hour per week).

• AI models could help negotiators secure peace deals (The Economist)

  • the Ukraine-Russia Peace Agreement Simulator. Users enter preferences for outcomes grouped under four rubrics: territory and sovereignty; security arrangements; justice and accountability; and economic conditions.

  • The AI model then cranks out a draft agreement. The software also scores, on a scale of one to ten, the likelihood that each of its components would be satisfactory, negotiable or unacceptable to Russia, Ukraine, America and Europe.

  • The model was provided to government negotiators from those last three territories, but a limited “dashboard” version of the software can be run online by interested members of the public.

• Un compte Tik Tok original qui utilise l'IA pour réimaginer des moments historiques vécus à travers l'oeil de quelqu'un "Point of View"

  • Une journée au Moyen-Âge, POV : 24h durant l'éruption de Pompéi, une journée lors de la Peste Noire, POV : 24h durant l'évasion d'Alcatraz

  • Beaucoup d'erreurs et de clichés, mais très original

• The Hypershell Exoskeleton Is So Good at Climbing Cliffs, It Ruined My Workout (source)

• MIT researchers have unveiled a remarkable robotic insect that may make pollination possible even without bees. (source)

  • These tiny robots, weighing less than a gram, mimic the flapping wings of insects and can hover for an impressive 17 minutes (over 100 times longer than earlier prototypes).

  • Imagine multi-level indoor farms where swarms of robotic insects synchronize to pollinate crops, maximizing yields while reducing environmental impacts.

• Funny: we evolved feet from hands, and not the other way around (source)

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

👋👀🦉💡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” 👇

• 🔒I Recorded Everything I Said for Three Months. AI Has Replaced My Memory.

• 🔒AI can cheat and it doesn’t seem to get any better

• 🔒How America and Europe are losing badly today's Electrical Age against China

• 🔒Comment l'armée française compte intégrer l'IA dans ses processus

• 🔒Why we are are Self-Domesticated Apes

• 🔒 Why Kerosene did NOT save the Sperm Whale

Les dernières newsletters :

• 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

• Why China will win on humanoid robots; Sam Altman's 3 observations on AI; robot with 1,000 artificial muscles & more

• Les précédentes

Merci, et à bientôt !

Thomas