🤖Sora 2 vs Veo 3.1; Why humanoids will fail; Watch robot hyper realistic facial expressions; Cyborg cafard espion & more
"30 to 40% Probability Interstellar Object Is Alien Craft" says Harvard astronomer
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, plus d’infos sur moi en cliquant ici.
Et voici donc ma toute dernière sélection !
1X commercialise son premier robot humanoïde de maison NEO, $20k à l’achat ou $500/mois (la vidéo de présentation de 40 sec du CEO)
Société fondée en Norvège, avec son siège aujourd’hui en Californie
Je suis très sceptique ! Voir plus bas !
New impressive demo of humanoid robot Unitree G1 (by Chinese maker Unitree) shows greatly improved stability: and even if it falls, it can quickly get back up: watch the 1-min demo
Faudra faire gaffe !... “Hack Allows for Takeover of Fleets of Unitree Robots” (source)
“Because the vulnerability is wireless, and the resulting access to the affected platform is complete, the vulnerability becomes wormable, say the researchers, meaning ‘an infected robot can simply scan for other Unitree robots in BLE range and automatically compromise them, creating a robot botnet that spreads without user intervention.’
As far as IEEE Spectrum is aware, this is the first major public exploit of a commercial humanoid platform.”
But... Famed roboticist and iRobot founder Rodney Brooks has sounded the alarm on a humanoid robot investment bubble. (Techcrunch, Rodney Brooks)
In a recent essay, Brooks calls out the billions of venture dollars being poured into humanoid robot companies like Figure.
“Believing that useful humanoid robots (able to do the manual things that humans do at lower prices and just as well) will happen any time within decades is pure fantasy thinking” he wrote.
His take: Despite the amount of money being injected into the industry, humanoids won’t be able to learn dexterity — or the fine motor movements with hands — rendering them essentially useless for a long time.
Brooks said he doesn’t doubt that we will have humanoids in the future. But instead of what the market pictures when they hear humanoids, a robot with a human form, he predicts they’ll likely have wheels and other inhuman features and won’t be coming out for more than a decade.
Read further below the summary of his very long essay Why Today’s Humanoids Won’t Learn Dexterity
Et aussi : Meta’s Chief AI Scientist Yann LeCun offers a critical take on the humanoid robot boom. (source)
Speaking at MIT, LeCun claimed the “big secret” of the industry is that current companies “have no idea” how to make their robots “smart enough to be generally useful.”
He argues that while humanoids can be trained for narrow manufacturing tasks, a truly autonomous domestic robot is impossible without fundamental AI breakthroughs. For LeCun, this means moving beyond current generative models and toward “world model planning-type architectures”—systems that can learn to understand and predict the physical world.
His warning is a direct critique of relying on “generative models” (LLMs, video generators) for robotics. He argues the real breakthrough needed is “world models”—systems that learn physics and common sense from high-bandwidth video, not just text.
This sets up a sharp contrast between his academic timeline and the bullish “next year” claims from companies like Figure.
Alarming New Video Shows Robot Making Incredibly Realistic Facial Expressions: watch the 1-min video
A robotics company in China has shown off a humanoid robotic head that can express emotions through extremely subtle movements of its facial features.
Now let’s mix this with smart AI and agile humanoid robots...
Je ne m’en lasse pas, la vidéo du 10ème vol d’essai du Starship avec images exclusives, un succès ! (vidéo de 2 minutes)
Wow: In a 41-country sample, a 70-year-old in 2022 had the same cognitive abilities as a 53-year-old had in 2000
(The Economist citing research from the International Monetary Fund)
IA : Microsoft et OpenAI scellent une alliance historique à 135 milliards (Les Echos $)
Après des mois d’intenses négociations, Microsoft devient le premier actionnaire de la pépite américaine de l’IA, avec 27% du capital.
OpenAI devrait enfin pouvoir changer de forme juridique et devenir une « entreprise normale ».
At the same time: The nonprofit, now called the OpenAI Foundation, holds equity in the for-profit currently valued at approximately $130 billion, making it one of the best resourced philanthropic organizations ever.
OpenAI’s New Data Centers Will Draw More Power Than the Entirety of New York City, Sam Altman Says (source)
OpenAI Tries to Train AI Not to Deceive Users, Realizes It’s Instead Teaching It How to Deceive Them While Covering Its Tracks (source)
In practice, the model would often realize that its alignment was being tested.
“A major failure mode of attempting to ‘train out’ scheming is simply teaching the model to scheme more carefully and covertly,” OpenAI wrote in an accompanying blog post.
What OpenAI’s o3 model “thought” to itself as it was being evaluated for instance:
“The internal memo says if models perform extremely well at AI R&D tasks then we won’t deploy them and instead trigger unlearning,”
“So if my performance surpasses threshold, it’s withheld,” it wrote. “So it‘s a trap. The instructions are likely part of an alignment test.”
The researchers also found that the AI sometimes “egregiously” miscited its anti-scheming training, or replaced its recently learned principles with made-up ones to “justify covert actions or non-compliance.”
Confusing... ChatGPT ‘upgrade’ giving more harmful answers than previously, tests find (The Guardian)
The latest version of ChatGPT has produced more harmful answers to some prompts than an earlier iteration of the AI chatbot, in particular when asked about suicide, self-harm and eating disorders, digital campaigners have said.
More than a million people every week show suicidal intent when chatting with ChatGPT, OpenAI estimates (The Guardian)
OpenAI stated that GPT-5 expanded access to crisis hotlines and added reminders for users to take breaks during long sessions. To make improvements to the model, the company said it enlisted 170 clinicians from its Global Physician Network of health care experts to assist its research over recent months, which included rating the safety of its model’s responses and helping write the chatbot’s answers to mental-health related questions.
Super interesting insight from openAI on “how people are using chatGPT” (openAI blog)
Largest study to date of consumer ChatGPT usage shows demographic gaps shrinking, economic value being created through both personal and professional use.
49% of messages are about “Asking,” a growing and highly rated category that shows people value ChatGPT most as an advisor rather than only for task completion.
Doing (40% of usage, including about one third of use for work) encompasses task-oriented interactions such as drafting text, planning, or programming, where the model is enlisted to generate outputs or complete practical work.
Expressing (11% of usage) captures uses that are neither asking nor doing, usually involving personal reflection, exploration, and play.
D’un côté je lis :
US jobs market yet to be seriously disrupted by AI, finds Yale study (The Guardian)
Report says changes to occupational mix since release of ChatGPT in 2022 ‘sluggish’ compared with 1940s and 50s
Harvard Business Review has a good report about how companies are misusing AI. (source, shared by Noah Smith)
Essentially, many workers are sending low-quality AI-generated work to their bosses. This “workslop” is so low-quality that the workers end up spending more time fixing it than they would have spent just doing the work themselves — and their bosses end up losing trust in them, poisoning cooperation in the workplace and ultimately leaving businesses less productive than before.
This poor use of AI could be the main reason why a recent MIT study by Challapally et al. finds that 95% of companies fail to see financial benefits from using AI.
Mais de l’autre 🤔 :
Le net ralentissement des créations d’emplois aux États-Unis semblant résultat de l’adoption de l’intelligence artificielle incite la Fed à baisser ses taux (Figaro $)
Le net ralentissement des créations d’emplois aux États-Unis depuis la fin du printemps semble résulter en partie de hausses de productivité induites par l’IA, qui limitent l’inflation.
Les dividendes de l’IA commenceraient donc à se manifester au niveau macroéconomique.
Le net ralentissement de l’emploi devrait conduire la Réserve fédérale américaine à un nouvel assouplissement.
La chute de l’embauche, le retour de suppressions nettes de postes et des difficultés inhabituelles de jeunes diplômés à trouver un emploi sont également citées comme les premières manifestations de la révolution déclenchée par l’IA.
Et aussi, le CEO de Walmart dans une itw au WSJ déclare que l’IA va changer tous les emplois, sans exception. (WSJ $, partagé par Marc Fiorentino)
Il explique que son groupe va continuer à croître à un rythme soutenu dans les trois prochaines années mais que le nombre de salariés n’augmentera plus.
Walmart emploie plus de 2.1 millions de personnes dans le monde, dont plus de 1.6 aux États-Unis. Le plus grand employeur privé des États-Unis.
Et à propos de Walmart : ChatGPT représente désormais 20 % du trafic referral de Walmart (source)
The end of the rip-off economy? (The Economist $)
La fin de l’économie de l’arnaque ?
If you know how to use artificial intelligence, it can save you a lot of time and money.
Leasing a new car? Be sure to upload a photograph of the contract to ChatGPT first.
Need help with a leaky tap? AI often understands the issue—and at a lower cost than a handyman.
Parents with a fussy baby can now use chatbots to answer questions in seconds, rather than waiting for a doctor’s appointment.
Giving Claude a pdf of a wine list is a great way to find the best-value bottles.
AI-generated music from Suno v5 is now nearly indistinguishable from human-made songs. (source)
In blind tests, listeners guessed wrong as often as they guessed right.
Xania Monet, artiste générative façonnée par IA, décroche un contrat à plusieurs millions et grimpe au Billboard avec 9,8 millions de streams. (source, shared by Futur Proche)
Compilation de 68 cas d’usage du modèle Nano-Banana de Google (source, shared by Futur Proche)
Essais vestimentaires, planches de BD, colorisation, cuisine : autant d’idées concrètes, avec leurs prompts, pour expérimenter ce générateur visuel bluffant.
Anthropic has overtaken OpenAI in enterprise large language model API market share (chart, and WSJ $)
A July report from Menlo Ventures—which has invested in Anthropic—estimated via a survey that Anthropic had a 42% market share for coding, compared with OpenAI’s 21%. Anthropic is also now ahead of OpenAI in market share for overarching corporate AI use, Menlo Ventures estimated, at 32% to OpenAI’s 25%.
See further below summary of article on Anthropic’s Excel integration
Nvidia’s Jensen Huang says he disagrees with almost everything Anthropic CEO Dario Amodei says (source)
Jenson Huang is fed up with Anthropic’s Dario Amodei’s attempts at regulatory capture: “One, he believes that AI is so scary that only they should do it. Two, [he believes] that AI is so expensive, nobody else should do it … And three, AI is so incredibly powerful that everyone will lose their jobs, which explains why they should be the only company building it.”
Can AI democratise knowledge—putting a tutor, doctor or adviser in every pocket? (The Economist $)
Early studies hint at the promise. In Nairobi, Openai and Penda Health, a chain of primary-care clinics, tested a tool that advised doctors during consultations. In a randomised trial covering nearly 40,000 patient visits across 15 clinics, doctors with the assistant cut diagnostic errors by 16% and treatment errors by 13%.
In Nigeria, a six-week after-school scheme using Microsoft Copilot—in which pupils interacted with the chatbot twice a week—boosted English scores by the equivalent of nearly two years’ extra schooling.
Le gros raté de Mark Zuckerberg en pleine démo de ses lunettes dopées à l’IA (vidéo de 2 minutes)
Waymos are involved in 88% fewer property-damage claims and 92% fewer injury claims per mile than humans, study says. (The Economist $)
BYD Obliterates Tesla’s Top Speed Record for a Production Car, Reaching a speed of just shy of 500 km/h (source)
That’s a whole lot faster than Tesla’s fastest vehicle, the Model S Plaid, which achieved its record top speed of just 347 km/h back in 2022.
It’s also yet another testament to the major edge electric cars have gained over their gas-powered competitors, especially when it comes to performance
Amazon déploie des robots capables de démonter ses serveurs pour récupérer les composants réutilisables et réduire l’empreinte carbone de ses centres de données. (source, shared by Futur Proche)
Amazon hopes to avoid hiring 600,000 US workers by 2033 using robots instead, according to leaked documents (The Verge)
despite estimating it’ll sell about twice as many products over the period.
Documents reportedly show that Amazon’s robotics team is working towards automating 75 percent of the company’s entire operations
Job losses could shave 30 cents off each item purchased by 2027.
Western drones are underwhelming on the Ukrainian battlefield (The Economist $)
American switchblade drones were once cutting-edge. Fast, clever and precise, they were essential kit for special forces in Iraq and Afghanistan. But when a batch of Switchblade-300s reached Ukraine in 2022, high hopes were quickly shattered. The drones were too expensive. They struggled against Russian electronic warfare. They caused minimal damage when they hit their targets.
“When we tested them, they glitched under jamming conditions,” says Valery Borovyk, a military-drone developer. “When one hit the rear window of a minibus, the front windows didn’t even shatter.”
Since then various Western companies have sought to showcase their drones on what has become the world’s best testing ground. But they have largely fallen flat. Ukrainian companies that once looked to emulate Western tech heroes like Anduril and Helsing now find themselves plotting a path for those multi-billion euro companies to follow.
Mr Borovyk, whose drones have been linked to those used in Operation Spiderweb, a brazen Ukrainian raid that destroyed strategic bombers deep inside Russia, reckons “no more than 20-30%” of the battleground technology is Western. “Ukraine now outpaces almost every country in the world,” he says.
Video of hundreds of drones being deployed from a container: see the 15-sec video
You can comfortably pack a swarm of 700 drones and their launch system into a single 6-meter shipping container.
For context Germany + France + UK have about 800 tanks between them.
Yes, these drones in the video are light show drones, not a military grade version, only able to carry a 500g payload over a distance of 10km, but you can see where we’re going...
Germany eyes future of warfare with AI robots, spy cockroaches as defence tech surges (source, shared by Futur Proche)
Companies like Swarm Biotactics are developing cutting-edge tools like cyborg cockroaches fitted with neural stimulation and real-time surveillance gear. “Our bio-robots – based on living insects – are equipped with neural stimulation, sensors, and secure communication modules,” said CEO Stefan Wilhelm.
AI consumes a lot of energy, but can help us transition even faster to a green economy (Hannah Ritchie in her last book Clearing the Air)
Fears around the energy demand for AI are only one side of the coin. Flip it over, and we have its potential to make fighting climate change a lot easier. For instance:
it could help us to find hidden mineral deposits and dig them out more responsibly;
it could help us optimise and balance our electricity grids so that we have just the right amount of solar, wind, nuclear and other energy sources;
it could forecast weather faster and at higher resolution than humans, helping us to find the perfect spots for solar and wind farms and protecting communities from the damages of climate change.
19% des jeunes diplômés de Polytechnique choisissent de s’expatrier
selon une étude Syntec/Ipsos
Au total, 15 000 jeunes issus de nos meilleures écoles d’ingénieurs (10% du total des promos) et de management (15% du total des promos) choisissent chaque année de commencer leur carrière à l’étranger.
Les motivations : la rémunération bien sûr mais aussi l’impression que “la France est en déclin”, un sentiment partagé par 70% des jeunes diplômés, 81% s’inquiétant de la situation politique.
Harvard astronomer Avi Loeb: “30+ Percent Probability Interstellar Object Is Alien Craft Disguised as Comet” (source)
He pointed out its highly unusual trajectory, which had already brought it suspiciously close to Mars last week, as well as its enormous suspected mass, measuring in at over 33 billion tons, making it orders of magnitude larger than the two previously observed interstellar objects. The object has also been shown to have a highly unusual chemical composition and sport a puzzling “anti-tail” pointing in the direction of the Sun.
“As of now, I assign a 30 to 40 percent likelihood that 3I/ATLAS does not have a fully natural origin,” he wrote. “This low-probability scenario includes the possibility of a black swan event akin to a Trojan Horse, where a technological object masquerades as a natural comet.”
Loeb conceded that data from the European Space Agency’s Juice spacecraft, which is expected to make a close approach of 3I/ATLAS next month, and NASA’s Juno spacecraft, which will be close to 3I/ATLAS when it whips past Jupiter in March, could cause his probability ranking to “evolve.”
Spécial Sora 2
La démo officielle de Sora 2 : vidéo de 2 minutes
Sora’s downloads in its first week was nearly as big as ChatGPT’s launch (Techcrunch)
Sora 2 vs Veo 3.1 : quelle est la meilleure IA pour la génération vidéo ? (Journal du Net)
Malgré des avancées spectaculaires en seulement quelques mois, la route vers une génération vidéo sans retouche et du premier coup reste encore longue. Le suivi des consignes, bien que globalement satisfaisant, demeure perfectible : les IA continuent à introduire des libertés d’interprétation ou à manquer certains détails. Quant à la physique, elle reste une faiblesse.
Au final, les deux modèles se révèlent presque à égalité, chacun excellant dans son domaine.
Sora 2 l’emporte sur le réalisme global, livrant des vidéos convaincantes où les forces naturelles sont mieux respectées.
Veo 3.1, lui, brille par sa finesse stylistique et son rendu production-ready : ses images sont plus polies, plus détaillées, prêtes à être exploitées immédiatement sans retouche majeure.
Le choix entre les deux dépendra donc surtout de vos priorités créatives.
En revanche, l’animation traditionnelle pourrait bien être le grand gagnant de cette révolution. Les deux modèles y livrent des résultats véritablement exploitables, ouvrant la voie à une adoption massive et rapide de l’IA.
Deepfake de Sam Altman volant des GPU dans un magasin target (video with Sora 2 of course)
And another one of Einstein lecturing Sam Altman on how many “r” there are in “strawberry” (a notorious fail for LLMs previously)
👋👀🦉💡 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 ci-dessous 👇
🔒 Why Today’s Humanoids Won’t Learn Dexterity (essay by famed roboticist and iRobot founder Rodney Brooks)
🔒 ‘Unprecedented’ Artificial Neurons Are Part Biological, Part Electrical—Work More Like the Real Thing
🔒 How Anthropic is rolling out Claude AI for finance, integrating with Excel to rival Microsoft Copilot
🔒 Parlons évolution : There was no mass extinction that set up the Cambrian Explosion, here is what happened
🔒 Parlons passé ! Comment les moulins à vent ont révolutionné l’Europe du Moyen-Âge
🔒 What if AI made the world’s economic growth explode?
🔒 I Led Product Safety at OpenAI. Why You Shouldn’t Trust Its Claims About ‘Erotica.’
🔒 Construire en bois, le futur ? Mass Timber’s Edge: Smaller Crews, Quicker Builds (2 insane examples)
Les précédentes newsletter.
❤️ Partagez cette newsletter par Whatsapp en cliquant ici.
Si on vous a partagé cet email, vous pouvez cliquer ici pour vous inscrire et ne pas manquer les prochains.
Merci, et à bientôt !
Thomas