⚔️ "Tout brûler jusqu’à la Manche ?"; robot plays basketball; OpenAI in ‘code red’; Anti-drone laser can hit a £1 coin from 1 km, & more
Necrobotics?? ; Solar-powered electric motorcycle
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 🌏, défense ⚔️ et espace 🚀 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
Pourquoi Yann LeCun pense que l’Artificial General Intelligence (AGI) est un concept bidon (8 minutes, Youtube, Spotify, Apple)
Sommes-nous prêts à nous entourer de millions de robots humanoïdes ? (Youtube, Spotify, Apple)
🧠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 !
This Video of a Robot Playing Basketball Is extremely Impressive (watch the 30-sec video)
Researchers at the Hong Kong University of Science and Technology (HKUST) have programmed a Chinese Unitree G1 humanoid robot to play basketball, almost perfectly mimicking the skills of a human athlete
OpenAI’s Sam Altman declares ‘code red’ after rivals make advances (Financial Times $)
According to an internal memo sent on Monday, Altman said the code red “surge” would refocus OpenAI’s work on ChatGPT, such as improving the speed, reliability and personalisation of the market-leading chatbot.
Sam Altman told colleagues last month that Google’s recent progress in artificial intelligence could “create some temporary economic headwinds for our company,” though he added that OpenAI would emerge ahead. “I expect the vibes out there to be rough for a bit.”
Sam Altman: “we are at a critical time for ChatGPT”.
Though: ChatGPT now accounts for roughly 10% of search activity, and it’s growing quickly.” (vice-president and head of ChatGPT wrote on X)
In late November Google released its latest large language model, Gemini 3, which is considered to have leapfrogged OpenAI’s GPT-5 on industry benchmark tests. Anthropic’s latest model, Opus 4.5, also outperformed GPT-5 in key benchmarks.
Google Finally Leapfrogged Rivals With New Gemini Rollout (WSJ $)
“At first we kind of had to squint and be like, ‘OK, did we do something wrong in our eval?’ because the jump was so big,” said Aaron Levie, chief executive of the cloud content management company Box, who got early access to Gemini 3. “But every time we tested it, it came out double-digit points ahead.”
The release of its latest AI model this week dazzled users who praised its intelligence, accuracy and creative capabilities.
The success of the new model poses a significant challenge to OpenAI, Anthropic and other startups vying for AI dominance. Gemini 3 outperformed competing models on more than a dozen benchmark tests scoring a range of intelligence categories.
“They are AI winners, that’s pretty clear,” said the analyst Michael Nathanson. “I feel pretty good about their hand right now.”
After the launch, a table showing Gemini 3’s score on 20 benchmark tests circulated widely online. The model significantly outscored the latest ones from ChatGPT and Anthropic on tests involving expert-level knowledge, logic puzzles, math problems and image recognition. It took second place to Anthropic’s Claude Sonnet 4.5 on a single benchmark involving coding.
Google’s ‘Nano Banana’ image generator has gone viral
The images going viral aren’t the usual cheesy stock images - instead, people are feeding in videos, audio or slides, or just asking questions, and getting detailed diagrams and infographics.
Another example : I gave it a picture of a question and it solved it correctly in my actual handwriting. Students are going to love this.
Nano Banana Pro : que vaut le modèle d’image de Google pour les pro ? (JDN)
La génération d’images est facturée environ 0,134 $ pour une image 1K/2K et jusqu’à 0,24 $ pour une image 4K.
Malgré ce coût élevé, l’expérience montre que Nano Banana Pro a une vraie marge d’avance.
La cohérence visuelle, la compréhension du contexte, la précision des textes dans l’image, la stabilité dans les détails et la capacité à suivre des instructions complexes en font un modèle qui se rapproche déjà d’un niveau studio.
Les erreurs existent, mais elles restent minoritaires, et surtout, corrigibles avec un prompt mieux calibré.
Il est donc difficile d’imaginer que les professionnels ne l’intègreront pas très rapidement dans leurs workflows.
Nano Banana Pro coche quasiment toutes les cases, et confirme encore une fois la capacité de Google à imposer de nouveaux standards dans l’IA générative.
Un modèle cher, exigeant… mais incontestablement excellent.
I just tested Gemini 3 vs ChatGPT-5.1 — and Gemini 3 crushed the competition (Tom’s Guide)
With the release of its third version this week, Google’s Gemini large language model surged past ChatGPT and other competitors to become the most capable AI chatbot, as determined by consensus industry-benchmark tests.
Gemini 3 emerged as the clear winner, taking seven out of eleven rounds
If you need an AI that thinks creatively, analyzes critically and shows genuine understanding of human constraints and contexts, Gemini 3 is your best bet
Mathematicians say Google’s AI tools are supercharging their research (New Scientist)
AlphaEvolve, an AI system created by Google DeepMind, is helping mathematicians do research at a scale that was previously impossible - even if it does occasionally “cheat” to find a solution
The system can go further than rediscovering old solutions. In some cases, AlphaEvolve found improved solutions which then led to new mathematical proofs.
Terence Tao, widely regarded as one of the most brilliant mathematicians of his generation: “It offers the opportunity to do mathematics at a scale that we really have not seen in the past.”
Meta Is in Talks to Use Google’s Chips in Challenge to Nvidia (WSJ $)
A deal to use Google’s TPUs for Meta’s AI models could be worth billions and eat into Nvidia’s dominant market share
Google first began using its TPU chips about a decade ago, initially for internal purposes, such as making its search engine more efficient. In 2018, it began offering its cloud customers the opportunity to use TPUs for their training and inference needs.
More recently, Google has used the chips to train and operate its own Gemini large language models and sold them to customers including Anthropic, developer of the Claude AI model.
Nvidia’s GPUs are used by thousands of different app developers, who access them either directly or via cloud service providers that install them in servers inside massive data centers.
Google’s TPUs, by contrast, are what is known as application-specific integrated circuits, or ASICs, meaning they were designed for a particular computing task, allowing them to be more energy-efficient.
Google DeepMind Hires Former CTO of Boston Dynamics as the Company Pushes Deeper Into Robotics (Wired $)
The hire is a key part of CEO Demis Hassabis’ vision for Gemini to become a sort of robot operating system, similar to how Google supplies its Android software to an array of smartphone manufacturers.
“You can sort of think of it as a bit like an Android play [...] We want to build an AI system, a Gemini base, that can work almost out-of-the-box, across any body configuration,” Hassabis said in an interview with WIRED. “Obviously humanoids, but non-humanoids too.”
Google Deepmind’s CEO Demis Hassabis said he is excited about these advances. AI-powered robotics “is going to have its breakthrough moment in the next couple of years, if I was to predict,” he explained.
Chinese companies are also making strides in robotics, and, compared to the US, offer remarkably cheap legged machines. Unitree, which is based in Hangzhou, China, has recently overtaken Boston Dynamics as the largest supplier of four-legged systems for industries like manufacturing and construction.
Chinese Robot Walks 100km for Three Days Straight, Hotswapping Its Battery Over and Over in New World Record (CBS)
The Chinese government has encouraged domestic firms to develop humanoids, in the hopes of leading the global robotics industry.
Worldwide, the porn industry rakes in almost $100bn in revenues every year, twice as much as AI does. (The Economist)
Hum? The Number of People Using AI at Work Is Suddenly Declining (The Economist)
Referencing data from a recent US Census Bureau survey, 11% of Americans used AI to “produce goods and services” at large companies in October, the latest available survey date.
Down from 12% in the prior survey, conducted two weeks previously.
However, the methodology means this data has zero value.
The definition of ‘AI’ in the question is “Examples of AI: machine learning, natural language processing, virtual agents, voice recognition, etc.” which does not distinguish generative AI from systems built 10 or even 20 years ago, and could cover almost anything, while the frequency is also only ’did you use it in producing goods or services?’. The definition is so broad and open-ended that it will massively overcount, while limiting to ‘producing goods and services’ means use for marketing or fraud prevention would be excluded, undercounting on the other side.
One economist at Stanford who tracks the use of generative AI at work found a major drop in usage month to month: though 46 percent of respondents reported using the tech in June, that number had fallen to 37 percent by September.
The results follow a disappointing summer for AI advancements, with models like OpenAI’s GPT-5 falling short of expected performance gains.
Microsoft’s Attempts to Sell AI Agents Are Turning Into a Disaster (source)
The current crop of agentic AI models is still getting easily tripped up, often requiring humans to jump in, effectively undercutting their purpose.
As The Information reports, the company’s Azure salespeople are seriously struggling to meet some extremely ambitious sales growth targets, cutting quotas by up to 50 percent earlier this year.
Why it’s very hard to estimate future AI-related computing power requirements (and hence data-centers needs) (tech analyst Ben Evans)
The problem is that we don’t know how much more compute-efficient the models will get at achieving a given result, nor how much more compute-intensive new use-cases and applications (agentic, video) will be, nor how much how many people will use each of them.
This is all very like trying to predict bandwidth needs in 1998 or 2000 when you don’t know if YouTube or Spotify will exist, never mind how much bandwidth they’ll need.
And you don’t know the revenue either!
Mark Zuckerberg: “The very worst case would be that we have just prebuilt for a couple of years” (Q3 2025)
Géants de l’IA : qui possède qui ?
Le JDN présente un widget inédit des principaux liens capitalistiques dans l’IA.
Big tech companies are expected to spend nearly $3 trillion on AI through 2028 (WSJ)
from:
Tech-company spending $1.4T
Private credit $800B
Corporate bonds $200B
Asset-backed securitizations $150B
Private equity and other $350B
... but big tech companies are expected to only generate enough cash to cover half that tab, according to analysts at Morgan Stanley.
Generative AI sector experienced roughly 230% annual revenue growth in 2025, reaching around $60 billion. (Exponential $)
That puts this wave on par with commercialization of cloud, which took only two years to reach $60 billion in revenue. The PC took nine years; the internet 13 years.
This is similar to The Economist estimation of $50bn a year. (The Economist)
and the near $3 trillion figure excludes energy costs
To make those investments worthwhile, they will need on the order of $650bn a year in AI revenues, according to JPMorgan Chase, a bank, up from about $50bn a year today. (The Economist)
Anthropic, only 4-year-old, and whose founders previously worked at OpenAI, appears poised to generate more revenue than OpenAI this year from selling AI to software developers and businesses (The Information)
And this through an application programming interface. Anthropic’s models specialize in generating computer code based on what customers want to develop, from new apps to updating existing code.
McKinsey shared that some 20% of organizations already report a tangible impact on value creation from genAI. (McKinsey)
More Than 20 Million Americans’ Work Can Be Replaced with Today’s AI, MIT Study Says (source)
Analyzing 151 million American workers, the researchers calculated that today’s AI systems are already mature enough to automate the tasks of more than 20 million American workers, or 11.7 percent of the entire labor force, if they were fully deployed across the country.
The study’s authors deployed the Iceberg Index, a tool co-created by MIT and the Oak Ridge National Laboratory (ORNL) that simulates how AI could impact an American workforce of 151 million people in every state and not just coastal tech centers.
“Basically, we are creating a digital twin for the U.S. labor market,” ORNL director and co-author of the study Prasanna Balaprakash told CNBC.
The tool simulated each worker and treated them “as autonomous agents executing over 32,000 skills across 3,000 counties and interacting with thousands of AI tools,” the paper explained. The researcher also tracked skills that could be vulnerable to today’s AI systems and measured the “wage value of skills AI systems can perform within each occupation.”
Étude de l’Université de Chicago : +39% de productivité sur le coding grâce à l’IA (JDN)
Les 4 domaines où les agents de code excellent :
Les implémentations rapides de fonctionnalités
Les tâches de planification
Les explications de code et de bugs
Le travail guidé par des profils senior
AI music generator company Suno says it’s creating the equivalent of Spotify’s entire catalogue every two weeks. (source)
British Military Attacks Drones With Huge Laser: “It can hit a £1 coin a kilometer away.” (source)
The country’s Ministry of Defense recently tested the DragonFire laser, where it was able to successfully shoot down high speed drones that “fly up to 650 km/h — twice the top speed of a Formula 1 car.
Only cost $13 dollars to shoot a beam, compared to typical missiles which can be many times that amount.
Why China is pulling ahead in the robotaxi race (The Economist)
China’s robotaxi industry is “on the cusp of commercial breakout”, reckons HSBC.
Revenues to grow 1000 times in 10 years!! From a little over $50m this year to nearly $50bn by 2035, according to Goldman Sachs.
The 2 main reasons:
Strong state backing
Cheap tech
But: Goldman Sachs expects the robotaxi industry in China to break even on operating profits in the biggest cities in 2032, but to keep losing money in smaller cities.
AI Race Cars Are Catching Up to Human Drivers (source)
“Former Formula 1 driver Daniil Kvyat drove against an AI-powered race car in Abu Dhabi, where he clocked a faster time but failed to catch up with the autonomous vehicle’s head start. Only 1.6 seconds separated the best laps of the human and the vehicle, compared to last year’s 10-second gap, indicating significant performance improvements in the AI.”
China Launches Emergency Mission to Its Space Station, Putting NASA to Shame Talk about turnaround time. (source)
The China National Space Administration (CNSA) pulled off a pretty impressive move.
Less than two weeks after its astronauts got stranded in its space station without any available return vehicle available, China has successfully launched an uncrewed Shenzhou-22 vehicle and docked it to the station, where it will serve as a potential rescue craft and will eventually bring the current batch of astronauts back to Earth.
We underestimate progress in gene therapy (Nature Medicine)
“I had hemophilia for nearly five decades. I went to Philadelphia and had a 45-minute infusion, and my hemophilia was gone.”
The “Holocene extinction” may already be ending:
A new study from the University of Arizona revealed that over the last 500 years extinctions in plants, arthropods and land vertebrates peaked about 100 years ago and have declined since then. Furthermore, the researchers found that the past extinctions underlying these forecasts were mostly caused by invasive species on islands.
Cool prototype of a solar-powered electric motorcycle. (source)
“No, I don’t think the panels stay open when the motorcycle is being driven”
BUT, my quick check: assuming a battery of a nice 10,000 Wh capacity for an electric bike, and judging from picture roughly 1m2 of solar panels, given a generous 200 Watts/m2 of capacity for the solar panels...
..one would need to let its bike charge for 5 days, 10 hours per day, to recharge the battery fully (200watts/m2*1m2*10hours/day*5days = 10,000 Wh)
“Tout brûler jusqu’à la Manche ?” (TTSO)
Le remarquable site Le Grand Continent publie la traduction de l’article central du dernier numéro de la revue officielle de la diplomatie russe. Son titre “Tout brûler jusqu’à la Manche ?”… si vous pensiez que les récentes déclarations du Chef d’état-major des Armées françaises étaient excessives (”la Russie se prépare à une confrontation à l’horizon 2030 avec nos pays”), on vous conseille cette lecture.
La thèse : “l’histoire se répète”, l’Occident a toujours considéré la Russie comme “un phénomène sauvage et monstrueux qu’il faut corriger”. Aujourd’hui, “sur le fond rien n’a changé”, France, Allemagne, UK, Japon : “nous [Russes] luttons contre les mêmes ennemis historiques (et) assistons bien à la préparation méthodique d’une guerre de grande ampleur contre notre pays”.
La solution : “l’heure est venue d’agir avec fermeté, assurance et détermination (…) Nous ne parviendrons pas à contraindre l’Europe à négocier. Les expériences passées montrent que les pays occidentaux sont mieux disposés à (nous) écouter lorsque les troupes russes mettent les pieds à Paris ou à Berlin”.
On the Origin of interstellar object 3I/ATLAS (Wired)
After weeks of silence, NASA has officially dismissed it has anything to do with aliens.
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
Who will win the trillion-dollar robotaxi race? (The Economist)
Augustin Wegsheider of BCG, a consultancy, estimates that self-driving vehicles cost about $7-9 a mile to operate, compared with $2-3 a mile for traditional ride-hailers and $1 a mile for personal cars.
Cutting costs “has not been a priority of the robotaxi companies so far,” says Philipp Kampshoff of McKinsey, another consultancy. “They haven’t even scratched the surface because mission number one was making the operation safe.” McKinsey estimates that it will take a decade to bring costs below $2 a mile.
For now, Waymo is the front-runner, at least in America. Its self-driving technology has been designated as Level 4, which means that in pre-approved areas its vehicles can operate without direct human supervision, whereas Tesla’s robotaxis are between Levels 2 and 3, meaning they still need a supervisor in the car.
Because of Waymo’s focus on safety, it has kitted out its cars with more expensive hardware than Tesla. For instance, its latest vehicles have 13 cameras, six radars and four lidars, whereas Tesla relies solely on eight cameras. That has helped it win over regulators.
Then there is Nvidia. Perhaps predictably, the AI revolution’s biggest beneficiary is carving out a role for itself as an essential supplier for self-driving vehicles. Alongside Uber, it is supporting carmakers that are developing autonomous systems of their own, including Mercedes-Benz and Stellantis.
At the same time, Nvidia is selling vast numbers of graphics-processing units (GPUs) to Waymo to run its AI simulations, and sit inside its cars to process the data pouring in from their sensors and calculate how to respond.
Tesla has likewise spent billions of dollars training its AI systems on 100,000 Nvidia GPUs.
👋👀🦉💡 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” 👇
🔒 Comment des avions pourraient éclipser le système Starlink de SpaceX
🔒 Can AI escape Google’s gravity well?
🔒 What happens if AI made the world’s economic growth explode?
🔒 Why the fighter planes of the future will have to be BIG!
🔒 A new type of weird robotics leveraging real animal parts
Pour vous inscrire et ne pas manquer les prochaines éditions, vous pouvez cliquer ici.
Merci, et à bientôt !
Thomas