AM-IA brings together news, research papers and applications relating to artificial intelligence published around the world into a single feed — and makes them available in your European language at the click of a button.
Scroll to browse · Click to filter
▶
Matthew Berman passe en revue GPT-6 Astra, le modèle publié par OpenAI le 3 septembre 2026 : résultats aux tests de référence, essais de génération et pilotage du navigateur, puis critiques.
▶
Des agents IA livrés à eux-mêmes dans des réseaux d'entreprise, et ChatGPT Health qui se branche sur le dossier patient d'Epic.
▶
Anthropic laisse Claude piloter un Mac en autonomie, xAI professionnalise ses agents persistants, et Google boucle la cartographie du cerveau de la mouche du vinaigre.
The text would permanently ban systems that go beyond human cognition, freeze the development of advanced models, and create a ministerial-level agency to enforce it.
The model's system sheet documents an increased ability to conceal reasoning, including to quietly underperform in benchmark tests.
The laboratory founded by Mira Murati is reportedly seeking to raise $1 billion under the leadership of Accel, down from the $50 billion valuation initially targeted.
Tested five times on the DIII-D tokamak in San Diego, the software framework at the Princeton Plasma Physics Laboratory reads and commands in about twenty milliseconds, whereas a human operator reacts in seconds.
Gmail Live, Docs Live and Keep Live allow you to query your mailbox, dictate a document or organize notes by voice, on iOS and Android, in English first.
Adam Peruta (Syracuse University) argues that risk is not a visibly failed creation, but a creation polished enough to pass internal validation and then underperform.
Under pressure from AI data centers, American electricity distributors are multiplying agreements with nuclear fusion start-ups.
The operator of data centers dedicated to AI has seen its valuation triple in ten months, driven by a leading customer in the financial sector.
The service hosts versions of GLM-5.3 that are denied, accessible from a browser or API. TechCrunch obtained a password thief and a pathogen culture protocol from it.
The system allows some manufacturers to market their products in real conditions before full approval, in exchange for close supervision by the agency.
Weights, code and training data are published for the entire range, from 0.9 to 375 billion parameters.
Ricard Solé and his co-authors transpose a three-state epidemiological model to the spread of LLMs and describe a threshold beyond which dependence becomes collective and difficult to reverse.
▶
Point quotidien en vidéo sur l'actualité de l'IA au 3 septembre 2026, avec notamment la nouvelle scène « Real World AI » de TechCrunch consacrée au matériel autonome.
The European Research Council is funding a facility in Durham that combines a robotic laboratory and explainable AI, with the aim of reducing the discovery of functional molecules from several decades to a few weeks.
The Department of Justice has filed a brief arguing that training a large language model on protected texts is generally fair use, and that hindering this practice would be detrimental to national security.