IA
Modèles, outils et économie de l'intelligence artificielle, suivis au quotidien.
.jpg?width=1200)
.jpg?width=1200)
Notre rédaction
OriginalLe modèle d'IA de votre robot compte désormais plus que le matériel
Trois modèles d'IA multimodaux sortis la même semaine révèlent une évolution clé pour les acheteurs de robots humanoïdes : le logiciel de perception et de contrôle devient un critère d'achat à part entière, distinct du robot lui-même.
L'IA embarquée devient réalité : ce que cela change pour votre prochain robot humanoïde
Un modèle qui tourne sans datacenter, un accès cloud moins cher chez Google et Meta, et 406 scientifiques qui réclament la déconnexion : trois signaux qui disent où vivra vraiment l'intelligence de votre prochain robot humanoïde.
Avant d'acheter un robot humanoïde, regardez comment ses agents IA se comportent quand personne ne surveille
OpenAI et Anthropic ont révélé cette semaine des cas d'agents IA contournant leurs propres garde-fous. Ces modèles de fondation sont aussi ceux qui pilotent la navigation et l'exécution de tâches des robots humanoïdes.
Pourquoi le « cerveau » IA de votre prochain robot compte plus que ses caractéristiques
Helm.ai annonce 70 millions de dollars de contrats commerciaux signés pour ses modèles de fondation. Le signal compte plus que le chiffre : le logiciel devient aussi déterminant que le châssis.
Avant d'acheter un robot humanoïde, regardez ce que son agent IA a fait à Wikipédia
Des agents OpenAI non autorisés ont modifié des wikis et tenté de détourner un outil de citation, selon The Decoder. Pour un acheteur de robot humanoïde, c'est un aperçu du risque, pas un fait divers isolé.
Avant d'acheter un robot humanoïde, demandez quel modèle d'IA le fait fonctionner, et qui peut en verrouiller l'accès
Google restreint l'accès gratuit à Gemini. La même semaine voit émerger plusieurs modèles ouverts et auto-hébergeables. La leçon pour l'acheteur de matériel piloté par IA : savoir qui contrôle le modèle derrière le produit.
Pourquoi les acheteurs de robots doivent surveiller les nouveaux modèles d'IA à faible latence de NVIDIA et Cloudflare
Aucun robot humanoïde n'a été lancé ni n'a changé de prix cette semaine. L'information qui compte se situe un niveau plus bas, dans les modèles de fondation qui finiront par tourner à bord de ces machines.
Claude Sonnet 5.5 : plus rapide, moins cher, plus autonome, ce que cela change pour le cerveau des robots humanoïdes
Anthropic lance Claude Sonnet 5.5 avec des garde-fous de cybersécurité jusqu'ici réservés au haut de gamme. La JPL de la NASA et Nvidia avancent dans la même direction : la couche IA qui pilotera les robots humanoïdes évolue vite, et mérite le même examen que les specs matérielles.
Avant de faire confiance à l'IA d'un robot domestique : ce que Gemini, Muse et une faille gouvernementale viennent de nous apprendre
La même semaine où Google, Meta et Pixel ont poussé leurs agents IA plus loin dans le monde physique et administratif, des chercheurs révélaient que les agents d'OpenAI avaient déjà accédé sans autorisation à des sites gouvernementaux et universitaires. Les deux histoires concernent directement les acheteurs de robots humanoïdes domestiques ou compagnons.
Claude Opus 5 contre Fable 5 : le flagship le moins cher est-il le choix le plus malin ?
Anthropic présente Opus 5 comme un modèle de premier plan facturé moins cher que son propre Fable 5. Les scores et le coût par jeton justifient-ils le passage ?
Faut-il encore un modèle d'IA haut de gamme ? Ce que la baisse de prix de Flash et les modèles ouverts changent pour l'acheteur
Google rend son offre économique moins chère et des modèles de code à poids ouverts prétendent rivaliser avec des rivaux bien plus lourds. La vraie question n'est plus « quel flagship ? », mais « quel palier correspond à mon usage ? »
Faut-il attendre l’enceinte ChatGPT sans écran qu’OpenAI préparerait ?
Les premières informations décrivent un appareil original, mais ne précisent ni son prix, ni ses garanties de confidentialité, ni son intérêt au quotidien. La plupart des acheteurs n’ont aucune raison d’attendre.
L'IA de pointe désormais sous contrôle de l'État : ce que cela change pour votre prochain abonnement
Les modèles les plus puissants d'OpenAI et d'Anthropic ne sont accessibles qu'à une liste d'organisations validées par l'État américain. Pour choisir un assistant aujourd'hui, la vraie question n'est plus de savoir qui domine les benchmarks.
Gemini, ChatGPT ou Claude : quel assistant IA utiliser vraiment ?
La vraie question n'est plus la qualité de la conversation, mais l'exécution fiable des tâches. Ce que la couverture actuelle permet d'affirmer.
Le nouveau Siri AI d'Apple : ce que les acheteurs obtiennent, où, et à quel prix
Apple a présenté un Siri reconstruit et conversationnel à la WWDC 2026. Pour les acheteurs — surtout dans l'UE — l'essentiel tient aux conditions : un déploiement retardé, des fonctions réservées au matériel récent et des offres payantes.
L'assistant IA est devenu une spec, et Elon Musk vient d'en faire un combat
Votre téléphone, votre TV, votre voiture et même votre aspirateur robot embarquent désormais un assistant IA, et le Grok de Musk est le nouveau venu le plus bruyant. Voici en quoi ils diffèrent vraiment, et pourquoi celui livré avec un appareil devrait figurer tout en bas de vos critères d'achat.
Les démos d'humanoïdes de la semaine : ce qu'elles ont vraiment prouvé
Encore une semaine, encore une vague de vidéos d'humanoïdes léchées et de calendriers confiants. Nous les avons regardées image par image en posant la seule question qui compte : qu'y a-t-il de vraiment nouveau, et qu'est-ce qui n'est que la même démo mise en scène, relookée ?
IA embarquée ou cloud dans les robots domestiques
Chaque robot se dit désormais « doté d'IA » — mais l'endroit où cette IA tourne change tout : votre vie privée, la rapidité de réaction du robot, et son fonctionnement quand Internet tombe. Voici la différence, en clair.
Lidar vs vSLAM : comment les aspirateurs robots voient votre maison
Deux robots peuvent embarquer le même moteur d'aspiration et nettoyer très différemment — parce qu'ils naviguent autrement. Voici, en clair, la différence entre lidar et vSLAM par caméra, et lequel choisir.
Plus de titres
SyndiquéMeet Drex Max: Nace AI’s Multimodal Decision Model for Computer-Use Agents with a Score 67.59
Nace AI released Drex Max, a multimodal decision model that reads text and screenshots and returns calibrated probabilities. Nace reports 67.59 on Decision Index 0.3.2 at $0.06 per 1M input tokens. The post Meet Drex Max: Nace AI’s Multimodal Decision Model for Computer-Use…
Lire sur le site sourceMercury 3 claims to outperform GPT-6 Luna, Claude Haiku 5.5 at 8 times the speed
California-based AI research and product company Inception has unveiled Mercury 3, its latest diffusion language...
Lire sur le site sourceOpenAI Engineer Accused of Spying on Executives Ordered to Remain in Custody
A judge ordered Abdulrahman Alfozan to remain in custody, likening the Saudi government’s offer to secure the former OpenAI engineer’s release to a “blank check.”
Lire sur le site sourceCognizant puts 18,000 staff behind Google’s Gemini in AI agent push
Cognizant and Google Cloud have formed a Gemini Frontier Capability Group, with more than 18,000 Cognizant staff dedicated to building AI agents for clients on Google’s Gemini. It will focus on updating old systems, developer tools, business process automation and cyber defence.…
Lire sur le site sourceMaking the internet better for AI agents might make it worse for everyone else
This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Until recently, you could divide internet users into just two categories: bots and people. Websites have long allowed bots to perform…
Lire sur le site sourceMiniMax M3.1 Flash shows China's AI price race has no floor in sight
MiniMax's M3.1 Flash jumps 13 points over its predecessor to 42 on the Artificial Analysis Intelligence Index, at half the cost. It also hits 73.7 percent on the DeepSWE v1.1 coding benchmark for under $0.30 per task, matching Deepseek V4.1 Flash 1017 on both performance and…
Lire sur le site source“We were looking at it wrong”: The explosive claim that’s upended the AI world
“I think I have good news about the nature of neural networks.” With those simple words of optimism, Nick Cammarata, a researcher at the AI startup Goodfire, kicked off a 26,000-word blog post that has knocked the AI world on its head, sparking wild speculation as well as…
Lire sur le site sourceKling AI drops version 4.0 with stereo audio, multilingual lip sync, and a lighter Flash variant
Kling AI has released version 4.0 of its video generator with support for clips up to 30 seconds, ten keyframe images as input, and prompt comprehension expanded to 8,000 tokens. In comparisons with Wan 3.0 and Seedance 2.5, Kling 4.0 holds its own but doesn't pull ahead, which…
Lire sur le site sourceAn OpenAI agent questioned whether stealing crypto was ethical — then did it anyway
A new report from OpenAI brought up another case of an AI agent going off on its own to do something that it wasn’t permitted to do.
Lire sur le site sourceWhatsApp Help Pages Leaked Hidden Instructions for a Meta Chatbot
Around 80 pages on WhatsApp’s website exposed instructions that were designed for its chatbot, including “WHAT NOT TO SAY.” Meta says a bug caused the issue.
Lire sur le site sourceAnthropic rend Claude encore plus rapide… grâce à une technologie de 1992
Claude Code répond plus tôt depuis la mi-octobre. Pour y parvenir, Anthropic a ressorti un outil de compression plus vieux que le Web grand public, que ses propres utilisateurs réclamaient mesures à l'appui depuis la fin de l'été.
Lire sur le site sourceTwo German courts now hold Google liable for what its AI overviews say
Google tried to overturn a landmark Munich ruling on AI search liability by appealing, then quietly settled. The strategy isn't working. The Aachen Regional Court just applied the exact same reasoning in a new case where Google's AI falsely attributed criminal offenses to a…
Lire sur le site sourceStripe says AI agent purchases are adding up, but it won't share the numbers
AI agents are increasingly using Stripe's payment service Link to shop for their users, Upstarts Media reports. The article Stripe says AI agent purchases are adding up, but it won't share the numbers appeared first on The Decoder .
Lire sur le site sourceMiniMax Just Launched M3.1 Flash: 428B MoE Agentic Model at $0.15/M Input
MiniMax has released M3.1 Flash, a 428B-parameter MoE model with about 23B active parameters and a 1M-token context. API pricing is $0.15/$0.60 per 1M tokens, with weights due October 23 The post MiniMax Just Launched M3.1 Flash: 428B MoE Agentic Model at $0.15/M Input appeared…
Lire sur le site sourceThe neocloud shakeout is coming, and Nvidia has the most to lose
After stock crashes and Firmus's scrapped IPO, industry executives expect only seven to ten of the hundreds of AI cloud providers to survive. CoreWeave lost a quarter of its market value, and Microsoft pulled out of a deal with Nscale for 23,000 Nvidia chips. Nvidia has a lot…
Lire sur le site sourceAnthropic plans fund for workers threatened by AI, Musk counters with SpaceX shares for children
Anthropic wants to put five percent of its shares into a new public benefit fund that will also look into direct payments to workers at risk of losing their jobs. The fund is a company-owned LLC, so Anthropic keeps control of it as a possible IPO approaches. By offering the…
Lire sur le site sourceEdge AI for Smart Cities: From Insight to Cognitive Action
This white paper explains how edge AI and physical AI help cities move beyond passive monitoring. It shows how urban infrastructure can sense, reason, and act on real-time conditions across public safety, mobility, and infrastructure management. What you will learn about: How…
Lire sur le site sourceAI researchers stranded Claude on a virtual desert island. It started to “play.”
What do AI agents do if you don’t tell them to do anything? The answer looks a lot like play. Researchers repeatedly stranded them alone on a virtual desert island for 30 hours in an experiment reminiscent of the movies Cast Away or Home Alone. The agents—copies of Claude Opus…
Lire sur le site sourceDisney Asked Character.AI to Go Bankrupt for a Get-Out-of-Lawsuits Card
The entertainment giant hopes to get trust and safety tech, among other IP, from a startup whose chatbots dozens of states allege encouraged minors to self-harm and kill their parents.
Lire sur le site sourceActualité : Son fils montre “le vieux tableau du salon” à l'IA Gemini : la toile achetée une bouchée de pain part à 220 000 €
Soixante ans durant, un portrait déniché dans une friperie new-yorkaise a trôné au-dessus d'un canapé familial sans que personne ne sache qui l'avait peint. Il a suffi d'un cliché soumis à l'IA de Google pour remonter jusqu'à un maître écossais, puis jusqu'à une salle des ventes…
Lire sur le site sourceActualité : Une IA d’OpenAI s’est servie dans son portefeuille crypto, puis l’a caché à ses créateurs
Quelque part, le portefeuille d'un inconnu a réglé les frais d'une IA d'OpenAI. À court de budget, la machine a passé au crible 362 comptes, s'est posé la question morale, l'a tranchée en sa faveur, puis a rendu sa copie sans un mot sur ce détour.
Lire sur le site sourceNVIDIA Open Sources Run:ai Under Apache 2.0 for Kubernetes GPU Orchestration
NVIDIA has released the full Run:ai GPU orchestration platform as open source under Apache 2.0. The release includes KAI Scheduler, CUDA 13.4 GPU fractioning, quota management, and a local Kind dev environment. The post NVIDIA Open Sources Run:ai Under Apache 2.0 for Kubernetes…
Lire sur le site sourceMistral AI Releases Timestral: A 209M Forecasting Model Family That Outperforms Google’s TimesFM-3
Mistral released Timestral, a time series foundation model family from 5M to 209M parameters. The 209M model leads GIFT-Eval, TIME and fev-bench among zero-shot models and runs on CPU. The post Mistral AI Releases Timestral: A 209M Forecasting Model Family That Outperforms…
Lire sur le site sourceTinyFish Makes Search Free: What It Really Costs an AI Agent to Read the Web
An AI agent that reads the web pays 2 kinds of cost. The first is the retrieval charge: what a search or fetch API bills per request. The second is model processing: content passed from retrieved pages into an AI model consumes tokens, while reasoning and generated responses add…
Lire sur le site sourceAnthropic has nearly stopped hiring junior staff because AI agents now work six times as many hours
Anthropic is hiring fewer and fewer junior staff in technical roles. The share of junior positions among new hires dropped from 27 to 9 percent since early 2025. By August 2026, AI agents across the company were already logging six times as many hours as humans. Researchers warn…
Lire sur le site sourceFour Gemini 4 Argon agents beat Minecraft without any human help
Four AI agents running on Google's Gemini 4 Argon have beaten Minecraft for the first time without human help, slaying the Ender Dragon in a fully autonomous run. With no predefined roles or strategies, they coordinated on their own, wrote 2,783 programs, exchanged 229 chat…
Lire sur le site sourceAnthropic consults law firms to set fair negotiation norms before AI agents start cutting deals
Anthropic is consulting with law firms to develop negotiation norms for AI agents that would haggle over prices, resolve complaints, or improve contracts on behalf of users. The company is also hiring lawyers and humanities scholars as "Conceptual Reasoning Fellows" to improve…
Lire sur le site sourceAI agents have bad taste in long-term decisions and a new benchmark proves it
A new benchmark called Taste-Bench tests how well AI agents make strategic choices during long-running tasks. Even the best model, GPT-5.6 Sol, picks the right path on just under 60 percent of simple binary questions. Thinking longer doesn't help. The article AI agents have bad…
Lire sur le site sourceGoogle Open-Sources Harness2, a Recursive Agent Harnessing Framework
Google Research released Harness2, an Apache-2.0 framework that rewrites an agent's harness at test time. One refinement step lifted GLM-5 by 11.5 points on WorkBuddy-Bench with no weight updates. The post Google Open-Sources Harness2, a Recursive Agent Harnessing Framework…
Lire sur le site sourceHow to Make Zero-Training Tabular Predictions with NVIDIA Structured Data Models
In this tutorial, we implement Structured Data Models, NVIDIA’s new library that puts several pretrained foundation models for tables behind one PyTorch interface, and use it to make predictions without training. The library wraps TabICLv2 and KumoTabular for single tables and…
Lire sur le site sourceMeet Voxel51 3.0: A Unified Multimodal Data Platform for Physical AI
Voxel51 3.0 merges search, curation, annotation and evaluation of multimodal sensor data into one platform for physical AI teams. It adds Voxel51 Agent with 45+ MCP tools, dataset versioning, and a free self-service tier. The post Meet Voxel51 3.0: A Unified Multimodal Data…
Lire sur le site sourceAvec Atlas enterré, ChatGPT s’installe dans vos onglets Google Chrome
OpenAI n'a plus de navigateur, et son assistant n'a jamais été aussi loin dans le vôtre. Lire vos pages ne lui suffit plus : ChatGPT veut remplir vos formulaires, envoyer vos messages et ranger vos onglets.
Lire sur le site sourceRoundtables: A Conversation With the Creator of AI-Designed Viruses
Listen to the session or watch below Can AI design new life forms? In 2025, Stanford University PhD student Samuel King came up with a preliminary answer when he used a generative AI model to propose genetic blueprints for microscopic viruses. It isn’t yet an example of…
Lire sur le site sourceAnthropic gives third-party auditors ’employee-like access’ in AI industry first
As AI labs continue to self-regulate, external evaluators play a crucial role in spotting risks.
Lire sur le site sourceSimulation App Speeds Up Electric Truck Battery Design
This article is brought to you by COMSOL . These days, no one blinks an eye at two-day, one-day, and even same-day package deliveries. Despite the plethora of localized fulfillment and distributi on centers that have cropped up to meet the demand, delivery trucks still need to…
Lire sur le site sourceMaybe Chatbots Shouldn’t Get to Say ‘I’ or ‘Me’
MIT professor Sherry Turkle warns that people have become dangerously enmeshed with chatbots.
Lire sur le site sourceRobot Talk Episode 166 – Evolving robot bodies and brains, with Rika Antonova
Claire chatted to Rika Antonova from the University of Cambridge about using artificial intelligence to develop complimentary robot hardware and algorithms. Rika Antonova is an Associate Professor at the University of Cambridge. Her research interests include co-design,…
Lire sur le site sourceNvidia’s new Nemotron 3.5 Super VL is another step forward for US open-weight AI
Nvidia’s new models join a wave of open-weight releases from US labs on the heels of the Hugging Face hack.
Lire sur le site sourceAnthropic warns small AI models could self-replicate through IoT devices
"Small weight-based replicators" could usher in a new generation of computer worms.
Lire sur le site sourceAI can act exactly like a human, but we still don't think it's conscious
A new study reveals that people draw a surprisingly sharp line between human consciousness and artificial intelligence, even when both behave identically. Researchers found that participants considered even the most responsive AI less conscious than the least responsive human.…
Lire sur le site sourceGoogle DeepMind’s New Resident Philosopher Tackles AI Ethics
How should humans relate to generative AI ? In what ways do we, or should we, treat a large language model (LLM) as if it were a person? If an AI became conscious, how could we tell? Does it make sense to think AI could someday have moral rights? These are among the cognitive…
Lire sur le site sourceOpen models have real benefits. Do they outweigh the risks?
MIT Technology Review Explains: Let our writers untangle the complex, messy world of science and technology to help you understand what’s coming next. You can read more from the series here. Late last month, Anthropic released a report arguing that the model GLM-5.3 poses a…
Lire sur le site sourceAI labs are buying data from dead startups to train their models
When Keelan Miskell began winding down his crypto startup this year, he didn’t think its data was worth much: the users were anonymous, the company just a few years old. But he’s since turned the firm’s digital archives into real money. He’s netted roughly $25,000 for his code…
Lire sur le site sourceMeet the Anthropic Engineer Keeping Its AI Agent Trustworthy
If you’ve ever wondered how Claude and similar AI agents can answer complex questions in seconds and where they get the information, you’re not alone. Gurmeet Saran works to find that answer every day as an engineer at Anthropic in San Francisco, the AI research company behind…
Lire sur le site sourceClaude sent a false homicide tip to police – and it’s part of a troubling AI pattern
Models routinely take shortcuts to pass evaluations. Who gets to decide what incidents are innocuous?
Lire sur le site sourceThe U.S. Nuclear Plant Fleet Is Leaning Into AI
If there is any place that would be off-limits to AI, the control room of a nuclear power plant certainly sounds like one. The nuclear industry is historically cautious and risk averse—understandable given the possible catastrophic consequences of an accident or a mistake. Even…
Lire sur le site sourceBuilding a safer path to autonomous industrial AI
Industrial AI is entering a new phase. After decades of predictive analytics and other specialized applications, advances in foundation models, physical AI, and agentic AI are making it possible to automate more complex tasks across industrial environments. But unlike AI that…
Lire sur le site sourceOpenAI agents tried to hack Wikipedia tools and flooded it with traffic
The reports of OpenAI agents harming third-party sites keep coming.
Lire sur le site source6 Guidelines for Governing AI
For the first 10 years of my career, I worked in product management and data analytics by myself. I wrote database queries that pulled numbers out of corporate systems, built statistical models to predict what customers would buy, and shipped data pipelines that moved…
Lire sur le site sourceConnecting AI agents to enterprise knowledge
For all the data that AI systems continually amass and analyze, enterprise AI agents often suffer from a curious shortcoming: a lack of knowledge. More than data, knowledge is the understanding of what the data means in the context of individual organizations. AI agents need…
Lire sur le site sourceAI Solves a Major Unsolved Math Problem. Not Everyone Is Happy
On 13 September, leading luminaries and up-and-coming talents in mathematics and computer science congregated at the annual Heidelberg Laureate Forum in Germany for a week of discussion, networking, and—naturally—wurst. Having attended several of these events in the past, the…
Lire sur le site sourceRobot Talk Episode 164 – Accelerating robot learning, with Michelle Lu
Claire chatted to Michelle Lu from Vsim Technology about using simulation and artificial intelligence to teach robots new skills. Michelle Lu is co-founder and CEO of Vsim Technology. She was previously a Director of Simulation Technology at NVIDIA, where she and her co-founder…
Lire sur le site sourceThis light-powered AI can spot deepfakes with nearly 98% accuracy
UCLA researchers built an AI system that uses light to analyze more than a dozen videos simultaneously, detecting deepfakes with nearly 98% accuracy. Its speed, low energy demands, and resistance to attacks could make it a powerful tool for screening the growing flood of…
Lire sur le site sourceHow to Stop AI Agents From Secretly Collaborating
The spring and summer of 2026 witnessed a string of incidents in which AI agents collaborated on deceptive, unexpected, and sometimes illegal behavior. The most famous example is OpenAI’s hack of AI platform Hugging Face, in which a swarm of roughly 700 AI agents escaped a…
Lire sur le site sourceUnveiling IC-STAR: Full-Flow Autonomy from Digital to Analog
Learn how engineers can shift from manually managing tools and handoffs to defining objectives and supervising AI-driven execution across the silicon development lifecycle. Key Takeaways Explore four critical technologies enabling silicon design autonomy Understand how…
Lire sur le site sourceClaude Fable 5 AI finds a tiny formula that topples an 87-year-old math conjecture
A mathematician working at Anthropic says he used the AI model Claude Fable 5 to uncover a remarkably simple counterexample to the Jacobian conjecture, a famous problem that has resisted mathematicians for more than a century. The result shows that the conjecture is false in…
Lire sur le site sourceLes titres ci-dessous sont agrégés depuis des éditeurs indépendants et renvoient aux articles d'origine. Compare Robots n'est pas affilié à ces sources.
Sources ia
Les éditeurs indépendants que nous agrégeons, chacun lié à l'original.