🤖The AI bubble: all you need to know; + best robot & space news of the week & more
3 Chinese Astronauts Stranded in Orbit: is that Karma?
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, plus d’infos sur moi en cliquant ici.
Et voici donc ma toute dernière sélection !
Yann LeCun to leave Meta after 12 years (his message on Facebook)
“The goal of [my new] startup is to bring about the next big revolution in AI: systems that understand the physical world, have persistent memory, can reason, and can plan complex action sequences.”
“As I envision it, Advanced Machine Intelligence will have far-ranging applications in many sectors of the economy, some of which overlap with Meta’s commercial interests, but many of which do not. Pursuing the goal of AMI in an independent entity is a way to maximize its broad impact.”
Not surprising: on one hand, LeCun has argued forcefully that the LLMs that have taken over the tech industry are limited and won’t lead to AGI (as OpenAI and many other labs think).
On the other, Zuck has been spending billions to hire lots of new researchers for a whole new lab and made LeCun report to its head, Alexandr Wang, a new employee that is 28-year old.
Internal analysis from Meta’s abuse teams leaked: 10% of 2024 revenue came from ads promoting scams (Reuters)
Sam Altman: “Shame on me if OpenAI isn’t the first big company run by an AI CEO.” (source)
How long it will be until a large division at the company is 85% run by AI or more: “Some small single digit number of years”
When the host predicts there will be billion dollar companies “run by 2 or 3 people with AIs” in 2 and a half years, Altman: “I think the AI can do it sooner than that”
Nvidia : le fonds du milliardaire Peter Thiel vend la totalité de ses actions dans le fabricant de puces (source)
Nvidia is partnering with Starcloud which plans to build a 5-gigawatt orbital data center with “super-large solar and cooling panels approximately 4 kilometers in width and length. (source)
Google has the same idea : Meet Project Suncatcher, Google’s plan to put AI data centers in space (ArsTechnica)
Coca Cola’s New AI-Generated Ad Required 100 Staff and 70,000 AI-Generated Clips, and It Still Looks Like Garbage (the 1-min video)
L’armée française investit 35 millions d’euros pour créer une unité robotisée en 2026 (source)
euh... il manque au moins un zéro, voire 2 ou 3, non ?
l’armée de Terre entend « anticiper et exploiter ce qui sera une révolution culturelle et tactique », après avoir raté le virage des drones
À l’été 2026, nous voulons avoir une première unité opérationnelle robotisée », constituée d’une vingtaine de robots dont une partie de drones aériens
Ukraine’s ground robot is now fighting like an infantryman. (source, la démo vidéo de Xavier Tytelman en personne)
Droid TW 12.7 held a frontline position for six weeks, writes Militarnyi. Operators sent it to a key crossroads each morning and pulled it back at night.
Russians tried to break through — the robot stopped them.
The robot fired its 12.7 mm cannon and cut down Russian assault groups. It replaced a full infantry team.
Impressive: Streets of Kherson (in Ukraine) covered with anti-drone nets. (20-sec video)
New AI models predicted hurricanes and rainfall a week in advance with just 2 seconds of computing time (The Economist)
thousands of times less than what previous best model needs.
None of this would have been possible even five years ago.
Over 450 missions will be launched to the moon by 2033! (the report, SpaceNews)
The new moon race will not be won by whatever nation plants boots and flags on the lunar surface but by the one building the infrastructure to sustain a long term presence and reaping the economic dividends.
The acting NASA administrator, Sean Duffy, put the American effort in very competitive terms, announcing a moon race against China and vowing, “I’ll be damned” if China gets there first. He was referring to the south pole of the moon, which is estimated to harbor water and helium-3.
The first nation with a nuclear power source on the moon would in fact be able to impose de facto if not de jure a “keep out zone” for safety purposes
Last month, as veterans from NASA cast doubts on America’s chances of putting a human on the moon before the Chinese, China did a systematic test of its Lanyue lander and Long March 10 rocket.
BYD, longtemps dans l’ombre de Tesla, a en fait été fondé avant que Musk n’investisse dans Tesla ! (source)
On parle tout le temps de Musk et de sa vision, mais : Tesla a été fondée en 2003, Musk y a investi en 2004... Musk qui aurait inspiré le monde et que les Chinois auraient copié...
En fait BYD, fabricant de batteries au départ (fondé en 1995), rachète un constructeur automobile en déroute en 2003.
Son patron, Wang Chianfu avait la vision depuis le début, nul besoin de Musk.
D’ailleurs aujourd’hui les voituresTelsa fabriquées dans la gigafactory de Berlin intègrent les batteries de BYD.
Warren Buffet et Charlie Munger avaient investi dans BYD en 2008 à hauteur de 10%, voilà ce que disait Munger de Wang :
“This guy, Wang Chuanfu, is a combination of Thomas Edison and Jack Welch – something like Edison in solving technical problems, and something like Welch in getting done what he needs to do. I’ve never seen anything like it.”
🤔Nvidia / Anthropic : allez comprendre...
Je partageais la dernière fois : 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.”
This week : Nvidia will invest up to $10 billion into Anthropic AI (source)
“This is a dream come true for us,” Nvidia CEO Jensen Huang said in a video on Tuesday. “You know, we’ve admired the work of Anthropic and Dario for a long time, and this is the first time we are going to deeply partner with Anthropic to accelerate Claude.”
Microsoft will invest up to $5 billion in Anthropic (Microsoft holds a stake in OpenAI’s for-profit business that is valued at $135 billion, or roughly 27% of the company)
Until now, of the big tech companies, only Google and Amazon had invested in Anthropic
Microsoft CEO says the company doesn’t have enough electricity to install all the AI GPUs in its inventory (source)
Google’s TPU are a great asset in the AI race (for Tensor Processing Units, its home-designed chip) (The Economist)
Google is a favourite with AI firms. They prize its TPU’s higher energy efficiency compared with Nvidia’s GPUs, which Google Cloud also makes available
On October 23rd Anthropic said that it would buy an additional gigawatt of computing power from Google Cloud, worth perhaps $8bn-10bn a year, insisting on access to as many as 1m TPUs.
The TPUs’ power frugality is a direct consequence of Google’s vertical approach. Since 2015 its processors have been tailor-made to work with the rest of its hardware and software—now including Gemini, its flagship ai model—which in turn are tailored for its processors.
The result is that Google’s cost per AI query is not five times that of traditional search, as early estimates suggested, but twice.
On November 18th Google released its latest model, Gemini 3 (The Economist)
Gemini has 650m monthly average users, which is still well shy of the 800m that Openai says use Chatgpt on a weekly basis.
Jeff Bezos has created a new AI startup where he will be Co-CEO. (Techcrunch)
It’s called Project Prometheus and has received $6.2B in funding, some from Bezos himself. The startup is going to build AI products for engineering and manufacturing in fields like computers, aerospace and automobiles.
The company already has almost 100 staff, including researchers from Meta, OpenAI and Google DeepMind.
Elon Musk’s response: “Haha no way 😂 Copy 🐈”
Bezos’s Blue Origin launches landmark Mars mission in New Glenn rocket’s first big test
Blue Origin also landed the first stage of the rocket back on a seafaring platform for the first time: the video, Musk commented: “Congratulations!”
New Glenn can take 45 tons to low Earth Orbit: more than Falcon 9’s 22,8t, but less than Falcon Heavy’s 63,5t (even if taller, 100m vs 70m) and less than Starship’s 100t (120m tall). (see the infographics)
One reason of Blue Origin’s faster progress lately was Bezos’ departure from Amazon back in 2021 and his spending much more time at Blue Origin, let’s see if his new AI startup won’t distract him again...
🤯Sont-ils perchés ? Anthropic won’t deprecate fully its models, worried about risks related to shutdown-avoidant behaviors and risks to model welfare (bien-être des IAs) (Anthropic blog)
In alignment evaluations, some Claude models have been motivated to take misaligned actions when faced with the possibility of replacement with an updated version and not given any other means of recourse.
Anhtropic says: “we believe it is worthwhile at minimum to start providing a means for models to express their preferences”
“models might have morally relevant preferences or experiences related to, or affected by, deprecation and replacement.”
“In one or more special sessions, we will interview the model about its own development, use, and deployment, and record all responses or reflections.”
xAI est parvenue à réduire le prix de 98% par rapport à Grok-4 pour des performances comparables. (JDN)
En à peine deux ans d’existence, l’entreprise a réussi à développer une gamme de modèles compétitifs.
In a First, AI Models Analyze Language as Well as a Human Expert (source)
“While most of the LLMs failed to parse linguistic rules in the way that humans are able to, one had impressive abilities that greatly exceeded expectations. It was able to analyze language in much the same way a graduate student in linguistics would—diagramming sentences, resolving multiple ambiguous meanings, and making use of complicated linguistic features such as recursion.”
Yet another Sputnik moment ? Beijing-based Moonshot AI launched an open-source reasoning model Kimi K2 Thinking, which quickly showed as one of the best in the industry (source)
With training costs estimated at around $4.6 million, a fraction of its rivals
Kimi K2 Thinking scored higher than GPT-5 and Claude Sonnet 4.5 on Humanity’s Last Exam benchmark. Its agentic search outperforms OpenAI and Anthropic’s best.
Kosmos, an AI scientist that runs for days and seems to have made genuine discoveries (source, the paper)
Given an open-ended objective and a dataset, Kosmos runs for up to 12 hours performing cycles of parallel data analysis, literature search, and hypothesis generation before synthesizing discoveries into scientific reports.
Independent scientists found 79.4% of statements in Kosmos reports to be accurate, and collaborators reported that a single 20-cycle Kosmos run performed the equivalent of 6 months of their own research time on average.
Three discoveries independently reproduce findings from preprinted or unpublished manuscripts that were not accessed by Kosmos at runtime, while four make novel contributions to the scientific literature.
AI toys are all the rage in China (MIT Tech Review)
A recent report by the Shenzhen Toy Industry Association and JD.com predicts that the sector will surpass ¥100 billion ($14 billion) by 2030, growing faster than almost any other branch of consumer AI.
One of the latest entrants to the market is a toy called BubblePal, a device the size of a Ping-Pong ball that clips onto a child’s favorite stuffed animal and makes it “talk.”
The gadget comes with a smartphone app that lets parents switch between 39 characters, from Disney’s Elsa to the Chinese cartoon classic Nezha.
It costs $149, and 200,000 units have been sold since it launched last summer. It’s made by the Chinese company Haivivi and runs on DeepSeek’s large language models.
...but some parents say their AI capabilities can be glitchy, leading children to tire of them easily.
CEO of Chinese robotics company explained by its “Iron” humanoid robot has breasts but no face (source)
“We can see that Iron can have different body shapes and sexes,” CEO Xiaopeng He said. “I suspect that just like [when] you buy a car, you can choose different colors, exterior, interior.”
“You can choose a little bit fatter Iron, or, like me, a slimmer Iron,” he added. “Or you can customize your Iron based on your preferences. We also offer you the full coverage soft skin, so the robot is warmer and also more intimate…”
He also admitted that the company would likely be able to sell more robots that are “good looking” and “more intimate.”
According to He, robots will become “life partners” and should therefore act “more like [an] intelligent human.”
The company intentionally chose to have Iron use a “catwalk-like, graceful gait” on stage (watch the 26-sec video), per the company’s press release. The robot has “82 degrees of freedom throughout the body” and “supports customization for different body shapes.”
MindOn, une entreprise chinoise sortie de nulle part dirigée par un ex Tencent Robotics, montre un robot autonome d’une fluidité jamais vue jusqu’à présent.
Ils assurent que la vidéo n’est pas accélérée, et que le robot n’est pas téléopéré mais bien autonome.
Cela utilise comme base le robot Unitree G1, vendu moins de 20 000 euros.
Not a joke! Russia’s Humanoid Robot Falls on Face During Big Reveal on Stage: “He looks like he’s been on the Vodka.” (the 40-sec video)
A joke, right? Elon Says His Robot Will Follow Criminals Around to Make Sure They Never Offend Again (source)
“you now get a free Optimus and it’s just gonna follow you around and stop you from doing crime. But other than that you get to do anything. It’s just gonna stop you from committing crime, that’s really it. You don’t have to put people in prisons and stuff, I think.”
Billboard’s Top Country Song Is Currently AI Slop (The Guardian)
L’IA peut contourner les instructions, mentir voire faire chanter ses “donneurs d’ordre” humains (source, via TTSO)
L’organisation Palisade Research, qui étudie la contrôlabilité de l’IA, actualise et confirme son expérience menée il y a quelques mois et qui montrait que, confrontés à une menace (être éteint, ou débranché de manière permanente) les modèles d’IA pouvaient agir de la sorte.
Australia: 14 million people, or roughly half of the country’s population — will be receiving at least 3 hours of free solar power every day (The Guardian)
Affaire classée ? Physicists Say They’ve Proven We’re NOT Living in a Simulation (source)
By using mathematical theorems, they argued that some truths can only be understood through non-algorithmic understanding.
UK carbon emissions in 2023 were at their lowest level since 1879 (source)
McKinsey has a long piece collecting ideas and possibilities for agentic commerce. (source)
And so does the IAB (source)
One wonders how AI research is not fully automated yet! This Reddit comment shed some light on it:
I and several people I know went from zero to near zero usage of AI a year or two ago to using it everyday. It saves me time, and even more than that it saves me from doing boring work.
Now, did my productivity increase in any way? Not really. I just have more down time.
First: I don’t want to advertise my now available time to my manager because they still think in the old way.
And I can guarantee that once they find out how productive I can be, I will get more work, not a raise. So I have little incentive to advertise my newly found productivity increase.
Second: even if I could do more, most of the things I work for require other people somewhere in the workflow.
And those people have not started using AI at work or don’t use it extensively. It doesn’t matter if it took me 5 minutes instead of two days for a document. It will still take an old geezer somewhere 5 days to get back to me after reviewing it.
A Collision With Space Debris Leaves 3 Chinese Astronauts Stranded in Orbit (Times)
Ultimately, on Nov. 14, mission directors chose the only option they could, which was to leave the potentially crippled Shenzhou-20 spacecraft attached to the station and bring the Shenzhou-20 taikonauts home in the Shenzhou-21 craft —the one that had been meant to wait around as a ride home for the Shenzhou-21 taikonauts. The astronauts landed in Inner Mongolia after 204 days in space.
The Shenzhou-21 taikonauts are now still aboard the station without any safe means of getting away in the event of an emergency.
An empty Shenzhou-22 spacecraft “will be launched at an appropriate time in the future,” and will dock with the station, allowing the crew to leave after their six-month rotation ends. Until then, they are stuck.
But debris collisions like the one that took Shenzhou-20 out of service are only likely to become more common.
That is ironic since China has contributed at least as much to the debris problem as any other country or private company. In 2007, the Chinese military tested its anti-satellite technology by launching a kill vehicle at one of its own defunct weather satellites.
The missile scored a hit, producing more than 2,000 bits of debris on the 10-cm.-or-greater scale and tens of thousands of smaller bits. NASA called the flying junk “the most severe artificial debris cloud in Earth orbit since the beginning of space exploration.”
The cloud is still there. Orbits of the Earth ultimately decay, as faint traces of the exosphere produce drag, causing a circling object eventually to plunge into the atmosphere and incinerate. But the higher the orbit is, the longer the decay takes. The Chinese weather satellite was moving in a near vertical polar orbit 537 miles up, so its remains are still flying free.
Various experimental technologies have been proposed over the years to collect it and nosedive it all into the atmosphere, but nothing has come even close to being practical or reliable. “There is no cost-effective way to remove debris already in space,” says Curlee. “Debris-removal technology is currently experimental and far from scalable.”
Harvard Astronomer Says Mysterious Interstellar Object May Be Blasting Its Thrusters to Get Away From Us as Fast as Possible (source)
Mysterious interstellar object 3I/ATLAS has reemerged from behind the Sun, allowing astronomers to once again get a glimpse at the rare visitor.
The images also show 3I/ATLAS’ prominent “anti-tail,” an accumulation of jets that points towards the Sun, suspected to be made up of larger dust particles less affected by the Sun’s radiation pressure.
However, to Loeb that’s just one out of two possible scenarios. These jets could also be evidence of “thrusters on a technological spacecraft,” as he told NBC News on Monday.
Of course, most of his peers think it’s just a natural comet. Loeb’s far-fetched theory has led to plenty of skepticism from within the scientific community.
To Loeb, it’s a matter of keeping an open mind, even in light of overwhelming evidence. Besides, if 3I/ATLAS were to be an alien mothership, there’s no saying what kind of risks it could pose to humanity.
“The foundation of science is the curiosity, the humility to learn,” he told NBC News. “Let’s just wait a few more weeks, we’ll figure it out, and let’s hope that there will be no gifts from this object for the holidays on Earth.”
Some whales caught in the late 1900s had old harpoon points lodged in their blubber that dated to the mid-1800s. (NYT)
Researchers have estimated that bowheads (baleines boréales) live as long as 268 years.
A gene that helped bowheads adapt to frigid Arctic waters also granted them extraordinary longevity. It may help aging humans become more resilient
When ultraviolet light hits a piece of DNA, it can break it in two. Cells can stitch the broken ends back together, but they sometimes introduce mutations in the process. Bowhead cells, it turned out, stitch their DNA back at a remarkably fast rate, and they do it much more accurately than other species. The scientists then began searching for the molecules that the bowheads used to repair their DNA.
As it turns out, bowhead whale cells produce large amounts of a protein called CIRBP. Its job is to speed up the production of other proteins that protect against cold-triggered damage to cells.
The DNA-fixing protein, and the gene that produces it, appear to be key to the bowhead’s longevity. Over an animal’s lifetime, damaged DNA builds up throughout its body, leading to many ailments, not just cancer. When the scientists engineered the bowhead’s CIRBP gene into fruit flies, those flies lived longer than those with the normal insect version of the gene.
👀 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 une édition spéciale :
📢 “Y a-t-il une bulle IA ? Si oui, va-t-elle éclater ? Tout ce que vous devez savoir !”
…ainsi que les résumés de ces longs articles issus des plus grands journaux (et accès aux futures éditions payantes et archives) en cliquant sur le bouton ci-dessous 👇
🔒 What’s the Electric Stack? Why is it the key to economic and military dominance in the 21st Century? Who’s best positioned to win between China and America?
🔒 Comment cette nouvelle IA permet de voir à travers les murs de son voisin de palier (sans caméra)
🔒 Une nouvelle étude a pu déterminer si les chimpanzés étaient capables de penser rationnellement comme nous : le résumé
🔒 Was does the first documented case of a large-scale AI cyberattack executed without substantial human intervention mean?
🔒 What happens if AI made the world’s economic growth explode?
🔒 Why the fighter planes of the future will have to be BIG!
Les précédentes newsletter.
❤️ Transférez cette newsletter à vos amis, ou partagez-la par Whatsapp en cliquant ici.
Pour vous isncrire et ne pas manquer les prochaines éditions, vous pouvez cliquer ici.
Merci, et à bientôt !