🤖Why China will win on humanoid robots; Sam Altman's 3 observations on AI; robot with 1,000 artificial muscles & 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.
👋👀🦉💡 Avant de vous partager ma toute dernière sélection gratuite, je vous invite 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” 👇
🔒 The 2 domains where superintelligence may first emerge
🔒 How AI agents will outmaneuver salespeople by optimizing persuasion
🔒 Exemples concrets de comment les pompiers utilisent l’intelligence artificielle dans leurs missions en France
🔒 Will AI be an equaliser, helping the poorest performers catch up with the rest, or will it enable high-flyers to fly still higher ?
Check the latest findings from important new research
🔒 The 10 most significant Chinese inventions from the past 15 years
🔒 How drones came to rule the battlefield in the Ukraine-Russia war
L’apéro
A Private Space Mission Just Successfully Landed on the Moon for the First Time (Wired)
View of the Earth 🌏 from the Moon 🌕 above
With its Blue Ghost lunar module, Texas-based Firefly Aerospace has just achieved what no other private company, anywhere in the world, has ever accomplished: successfully landing on the surface of the moon.
The Blue Ghost lunar module lifted off on January 15 aboard a SpaceX Falcon 9 rocket and traveled nearly 5 million kilometers to reach its destination, at a cost of $2.6 billion. Among the instruments it is carrying are a lunar soil analyzer, a radiation-tolerant computer, and an experiment that tests the feasibility of using GPS to navigate the moon.
AI-video "If Trump was President in 1940..." with Zelensky replaced by Churchill: see the 90-sec video
hum...Elon Musk's Grok AI estimates with 75%-85% certainty that Trump is a Putin compromised asset. (source)
World's largest automated port (in Singapore) will handle 65 million containers per year (source)
For comparison, the Port of Marseille handles around 1.3 million TEU of containers annually, making it the largest container port in France
Demis Hassabis, CEO of DeepMind Technologies, calls AGI-level reasoning the ability to invent relativity with only the knowledge that Einstein had at the time.
OpenAI has released its successor: GPT-4.5. (NYT)
The new technology signifies the end of an era. OpenAI said GPT-4.5 would be the last version of its chatbot system that did not do “chain-of-thought reasoning.”
After this release, OpenAI’s technology may, like a human, spend a significant amount of time thinking about a question before answering, rather than providing an instant response.
GPT-4.5, which can be used to power the most expensive version of ChatGPT, is unlikely to generate as much excitement as GPT-4, in large part because A.I. research has shifted in new directions. Still, the company said the technology would “feel more natural” than its previous chatbot technologies.
“What sets the model apart is its ability to engage in warm, intuitive, naturally flowing conversations, and we think it has a stronger understanding of what users mean when they ask for something,” said Mia Glaese, vice president of research at OpenAI.
OpenAI said that, beginning Thursday, GPT-4.5 would be available to anyone who was subscribed to ChatGPT Pro, a $200-a-month service that provides access to all of the company’s latest tools.
It’s trained on an order of magnitude extra compute and thus an extremely expensive model (it’s priced at $75/$150 in the API vs GPT 4o’s $2.5/$10 price point).
But if history repeats, we could see 5x lower prices by year-end.
"This model seems to be somewhat better than the preceding 4o, though not dramatically so, but doesn’t match o1 and o3, and is 15-30x more expensive to run than 4o, which puzzled a lot of people" says tech analyst Ben Evans
He goes further: "OpenAI and all the other foundation model labs have no moat or defensibility except access to capital, they don’t have product-market fit outside of coding and marketing, and they don’t really have products either, just text boxes - and APIs for other people to build products."
Y Combinator Pulls Support for AI Startup After Video Emerges of Boss Barking at Human Worker, Calling Him "Number 17" (source)
Startup called Optifye, a Y Combinator-backed venture that's "building AI performance monitoring for factory workers, boosting line efficiency for manufacturing companies."
The surveillance platform caused a firestorm on the internet this week after a demo video surfaced showing a supervisor use the software to hone in on a worker who he referred to as "Number 17" instead of a human name, berating him for poor performance on the factory line.
According to the demo, Optifye represents each worker with a numbered rectangle, colored green if performance is up and red if performance is down.
Watch the 45-sec video reposted elsewhere after it was deleted by YC
In the same vein, look also at:
corporations like Walmart, Delta, Starbucks, and Chevron are already partnering with AI monitoring companies like Aware to surveil workers for thought crimes like talking about unions, wages, and working conditions.
Si c'est utiliser l'IA pour faire ça...vivement qu'on passe directement aux robots...
Mark Zuckerberg, Meta’s chief executive, stirred alarm among developers last month when he predicted that A.I. technology sometime this year would effectively match the performance of a midlevel software engineer, though he later suggested that it could free up human developers to be more creative. (source)
Microsoft CEO Admits that AI is far from generating as much value as the current hype would suggests "The real benchmark is: the world growing at 10 percent." (source)
Three Observations on AI (essay by Sam Altman, OpenAI’s CEO)
Systems that start to point to AGI are coming into view, and so we think it’s important to understand the moment we are in. AGI is a weakly defined term, but generally speaking we mean it to be a system that can tackle increasingly complex problems, at human level, in many fields.
In a decade, perhaps everyone on earth will be capable of accomplishing more than the most impactful person can today.
3 observations:
1. The intelligence of an AI model roughly equals the log of the resources (data and compute) used to train and run it.
log pour fonction logarithme, l'inverse de l'exponentielle : ici, plus on a de "data & compute", plus on a d'intelligence, certes, mais avec des rendements décroissants
2. The cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use.
3. The socioeconomic value of linearly increasing intelligence is super-exponential in nature.
A consequence of this is that we see no reason for exponentially increasing investment to stop in the near future.
Imagine it as a real-but-relatively-junior virtual coworker. Now imagine 1,000 of them. Or 1 million of them. Now imagine such agents in every field of knowledge work.
L'IA toujours plus une "commodity", 3 exemples :
Le chinois Kimi AI vient de lancer K1.5, un modèle de raisonnement supérieur à R1 (source)
K1.5 peut raisonner sur du texte, des images mais aussi des documents.
Tencent’s AI Bot (called Yuanbao AI) Passes DeepSeek as China’s Favorite on iPhones (Bloomberg)
Researchers in the US claim to have trained their s1 LLM for just $6. (The Economist)
Where DeepSeek took 2.7m hours of computer time to train; s1 took just under seven hours.
s1 is instead “fine-tuned” on the pre-existing Qwen2.5 LLM using a very small amount of data, of high enough quality: a streamlined 1,000 questions
Markets are not betting on AGI yet.
30-year US bonds still have low real interest rates (~1.6%). If investors believed in a near-future AI revolution, we’d expect different yields, since they should reflect expectations of future economic growth – so either the market is right, or there’s a potential trillion-dollar opportunity in betting they’re wrong. (source)
"Move 37" (by Andrej Karpathy, OpenAI cofounder and previous AI director at Tesla)
It's when an AI, trained via the trial-and-error process of reinforcement learning, discovers actions that are new, surprising, and secretly brilliant even to expert humans. It is a magical, just slightly unnerving, emergent phenomenon only achievable by large-scale reinforcement learning. You can't get there by expert imitation. It's when AlphaGo played move 37 in Game 2 against Lee Sedol, a weird move that was estimated to only have 1 in 10,000 chance to be played by a human, but one that was creative and brilliant in retrospect, leading to a win in that game.
We are seeing the first very early glimmers of things like it in open world domains. The models discover, in the process of trying to solve many diverse math/code/etc. problems, strategies that resemble the internal monologue of humans, which are very hard (/impossible) to directly program into the models.
Weird as it sounds, it's plausible that LLMs can discover better ways of thinking, of solving problems, of connecting ideas across disciplines, and do so in a way we will find surprising, puzzling, but creative and brilliant in retrospect.
It could get plenty weirder too - it's plausible (even likely, if it's done well) that the optimization invents its own language that is inscrutable to us, but that is more efficient or effective at problem solving.
I don't think we've seen equivalents of Move 37 yet. I don't know what it will look like. I think we're still quite early and that there is a lot of work ahead, both engineering and research. But the technology feels on track to find them.
About Ukraine's innovative powers (The Economist)
The drones over Ukraine are upgraded every few weeks, a pace that is beyond the Pentagon’s budgeting process, which takes years. American and European jammers in electronic warfare cost two or three times as much as Ukrainian ones, but are obsolete. Many big American drones have been useless in Ukraine; newer ones are pricier than Ukrainian models.
China has the world’s best delivery robots: Alibaba's Driverless Robots Make 10 Million Deliveries (source)
World's Largest Call Center Deploys AI to "Neutralize the Accent" of Indian Employees (Bloomberg)
Paris-based outsourcing company Teleperformance — which works with clients including Apple, Samsung, and TikTok — invested $13 million earlier this year in Sanas AI, a "real-time speech understanding platform" that boasts a so-called "accent translation" feature that uses machine learning to scrub the accents of overseas customer service workers.
Norway’s company 1X Is Building a Humanoid Robot for the Home (Techcrunch)
1X, which OpenAI was announced as an early backer of, has built the robot to be welcoming, with a friendlier design and a suit made of knitted nylon. The latter is designed to reduce potential injuries that might arise from robot-to-human contact.
Images of the robot show it performing a number of household tasks like making coffee, doing the laundry, and vacuuming."
Poland-based Clone Robotics released video footage of its Protoclone humanoid robot, a full-body machine that uses synthetic fluid-filled muscles to create unsettlingly human-like movements (source)
In the video, the robot hangs suspended from the ceiling as its limbs twitch and kick : watch the 40-second disturbing video
It has a polymer skeleton that replicates 206 human bones. The company built the robot with the hopes that it will one day be able to operate human tools and perform tasks like doing laundry, washing dishes, and preparing basic meals.
The Protoclone reportedly contains over 1,000 artificial muscles built with the company's "Myofiber" technology.
These muscles contain balloons that contract when filled with hydraulic fluid, mimicking human muscle function.
A 500-watt electric pump serves as the robot's "heart," pushing fluid at 40 standard liters per minute.
DeepSeek R1 vs OpenAI o3-mini : lequel est fait pour vous ? (JDN)
o3-mini high est ainsi le modèle le plus adapté pour les questions complexes, c'est aussi le meilleur modèle de raisonnement d'OpenAI publiquement accessible, la sortie d'o3 version "classique" étant définitivement abandonnée. C'est ce modèle que nous comparons aujourd'hui à R1.
Dans la théorie :
o3 serait donc plus pertinent pour les questions nécessitant un raisonnement mathématique ou plus généralement pour la génération de code et la résolution de bugs logiciels.
R1 serait plus adapté pour des questions de raisonnement généraliste ou des questions factuelles sur un domaine donné (hors mathématiques et code).
Dans la pratique, o3-mini et R1 se valent
Comparaison sur 3 cas d'usage précis : la génération de code, la découverte d'une vulnérabilité logicielle et enfin la recherche web avancée.
Sartup Ex-Human licenses its AI models to dating company Grindr, which is working on an “AI wingman” that will help users keep track of conversations and eventually may even date the AI agents of other users.
Large US Law Firm Sends Panicked Email as It Realizes Its Attorneys Have Been Using AI to Prepare Court Documents (Reuters)
Last week, a federal judge in Wyoming admonished two Morgan & Morgan lawyers for citing at least nine instance of fake case law in court filings submitted in January. Threatened with sanctions, the embarrassed lawyers blamed an "internal AI tool" for the mishap, and pleaded the judge for mercy.
A “true crime” documentary series has millions of views, but the murders are all AI-generated (source)
the videos' scripts were half generated by OpenAI's ChatGPT and half written by Paul, the author, and the visuals came from an AI image generator whose name he didn't disclose.
Paul was defensive about the "absurdist art form" he believes he pioneered.
"If people don't understand it, that says a lot about human nature and their own natures and the nature of crime," he told the website, "and perhaps they're not willing to question themselves."
Even the seemingly silly government-funded research projects that conservatives love to make fun of — shrimp running on treadmills! — often result in very practical discoveries. Some examples (source):
Research into frog skin secretions unlocked the mysteries of fluid absorption, paving the way for oral rehydration therapy. The payoff has been nothing short of remarkable: over 70m lives saved, most of them children in developing nations
Scrutinizing fly reproduction yielded an elegant solution to America’s livestock pest problem: the sterile screwworm fly. This peculiar triumph of entomology now saves ranchers $200m annually
When scientists first examined Gila monster venom (a lizard), they were hardly dreaming of treating diabetes and obesity. Yet this toxic cocktail held the key to developing GLP-1 agonists like Ozempic. Maybe you’ve heard of this drug?
YouTube says it now has 1bn podcast viewers. "The YouTube of podcasts turned out to be YouTube." said tech analyst Ben Evans (source)
This guy built a fusion reactor prototype for $3000 in his kitchen using Claude AI and parts ordered online... (source)
Parlons Passé
A message in a “bottle” from 2300 years ago: “ I, Gong Cheng De, together with Jiu Jiang Man, who are guards of the king’s tomb, hereby fishing on order of the king, sends our regards to future generations!” : The 19 characters in Chinese , is a stone inscription which was forgotten after writing for the next 2200 years until 1937, when a farmer in China founded and brought it home to be used as a tool. (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
China may take the lead on humanoid robots (Bloomberg)
Chinese robotics company Unitree has shown off its G1 humanoid robot pulling some gnarly kung fu moves.
In a crazy 25-sec youtube video released this week, the child-sized robot can be seen punching the air and even delivering a swooping roundhouse kick.
Unlike its growing number of Western counterparts, including Tesla's Optimus and Figure's 01, Unitree's G1 is already being sold to consumers. Pricing starts at just $16,000, according to the company's website.
versus the potential $20,000 to $30,000 Elon Musk may ask for Tesla’s Optimus robot in the future.
A separate even more impressive 1-min video released last month shows it running down a street and briskly walking over an uneven, rocky surface.
When Chinese President Xi Jinping gathered two dozen of the nation’s business leaders for a summit last week, one of the surprise attendees was a little-known, 34-year-old robot pioneer: Wang Xingxing, chief executive officer of Unitree Robotics
He was seated in the first row in front of Xi, more central than celebrated founders such as Alibaba Group Holding Ltd.’s Jack Ma and Tencent Holdings Ltd.’s Pony Ma. The country’s state media showered him with attention after the summit, which included a coveted handshake with the president.
Wang has said humanoid robots are evolving faster than even he expected, and such products may become widely deployed in service and manufacturing sectors by 2026 or 2027.
Unitree has the potential to do for China’s robotics sector what DeepSeek did in artificial intelligence. Wang and his team have pulled off technological breakthroughs at relatively low cost — with parallels to DeepSeek’s bombshell earlier this year — raising the potential for China to field throngs of robots for industrial, commercial and even military use.
Some American politicians have expressed concerns that such machines may threaten the country’s business and defense interests.
There were over 160 humanoid-robot manufacturers worldwide as of June 2024, of which more than 60 were in China, more than 30 in the United States, and about 40 in Europe. (MIT Tech Review)
In addition to having the largest number of manufacturers, China stands out for the way its EV sector is backing most of these robotics companies.
China’s extensive supply chain infrastructure supports these developments. According to a report by Morgan Stanley, China controls 63% of the key companies in the global supply chain for humanoid-robot components, particularly in actuator parts and rare earth processing.
This dominance enables Chinese manufacturers to produce humanoid robots at lower prices than their international competitors. Unitree’s H1 is priced at $90,000—less than half the cost of Boston Dynamics’ Atlas, a comparable model.
Les dernières newsletters :
L’addition ?
Cette newsletter est gratuite, si vous souhaitez m'encourager à continuer ce modeste travail de curation et de synthèse, vous pouvez prendre quelques secondes pour :
transférer cet email à un(e) ami(e) ou partager par whatsapp
et ici pour la version 100% en anglais
étoiler cet email dans votre boîte mail
cliquer sur le coeur en bas d’email
Un grand merci d'avance ! 🙏
Ici pour s’inscrire et recevoir les prochains emails si on vous a transféré celui-ci.
Merci, et à bientôt !
Thomas