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
▶
Tour d'horizon vidéo des nouveautés IA d'octobre 2026 par StudyZoom : GPT-6.1 Sol, les agents permanents Dots d'OpenAI, Claude Opus 5.5 et d'autres outils pour développeurs et pour la productivité.
Co-ordinated by the game’s chat function and a shared memory, with no assigned roles or hand-coded logic, four agents completed a survival game after 2,783 programmes had been written and 605 deaths.
By incorporating penalties for local maxima and gradients into the training process, researchers have improved the super-resolution of 150 out of 168 wind fields, with topological deviations reduced by between 25 and 31 per cent.
A test set comprising 300 resolved clinical cases evaluates simulated ‘consultation meetings’ between models. Structured reports outperform unstructured ones in 7 out of 8 pipelines, with GPT-5.6-Sol coming out on top.
The bank considers the valuation premium on AI-related shares to be unsustainable (23 times earnings compared with 13) and advises diversifying, whilst retaining infrastructure providers.
Appearing before the Advisory Committee on Rules of Evidence in Boston, the two groups concluded that the scientific detection of computer-generated images was doomed to failure. The judges remain cautious about amending Rule 901.
The new version of Anthropic’s code agent generalises sub-agents that inherit the conversation context, speeds up the loading of plugins and adds a compaction setting in the event of inactivity.
The two financial giants block unauthorised third-party agents: read-only access to accounts via MCP servers, but no transactions. Schwab blocks dozens of them every week.
The Canadian start-up has secured its first major US property developer: D.R. Horton will have more than 1,250 homes built via its ‘factory-to-home’ scheme, where AI translates plans into instructions for robots.
For Nvidia’s NVHBM memory, Samsung will manufacture the DRAM layers but will have TSMC etch the base logic chip using a 3 nm process, as it lacks sufficient in-house capacity and yield rates.
▶
Un agent chargé de réserver un restaurant a passé 200 appels en 90 minutes et fait bannir son utilisateur de Resy. Un panel débat des garde-fous nécessaires avant de confier une vraie autonomie aux agents.
The U-Net model used to clean the KaRIn altimeter measurements produced artefacts in conditions of heavy swell or in the presence of internal waves. A new training procedure corrects these biases.
Trained on five years’ worth of SuperDARN and DSCOVR data, three deep neural networks were compared: the probabilistic model performs better as the time horizon lengthens and the event intensifies.
A biology-based reasoning game compares three types of AI agent. When presented as a partner, the agent encourages children to ask it questions; when presented as a tool, it encourages them to look for evidence.
Rather than assessing an approach on the basis of a brief trial, the method predicts its long-term effectiveness. Across 108 simulated scenarios, it reduces the number of failed recoveries by 27.8 points.
By retraining a segmentation network using scans of patients with defibrillators, researchers have improved the delineation of 20 substructures of the heart despite metallic artefacts.
In an interview with Joanna Stern, the Microsoft boss outlined a version of Windows in which each agent has its own identity, monitored by Defender, and a $2,600 RTX Spark laptop capable of running large models without relying on the cloud.
A clause in Meta’s terms and conditions suggested that it could train its AI on any content from domains linked to a business account. It has since been removed.
The Swedish start-up, valued at $245 million, is using 18 years’ worth of cell testing data inherited from Wildcat Discovery to train its models, with a view to speeding up the development of new cathodes.
Google Quantum AI has demonstrated on its Willow chip that a filtering protocol produces quantum states with 99.99 per cent logical fidelity, overcoming one of the major hurdles to universal quantum computing.