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
▶
En 2020, Citibank a versé par erreur 900 millions de dollars aux créanciers de Revlon malgré trois validations. John Feminella (Two Sigma) en tire cinq garde-fous pour donner de l'autonomie aux agents IA.
By organizing data, tools and procedures in reusable bricks that improve in use, MatOS exceeds Codex CLI and Claude Code with an equal basic model on several scientific test benches.
Pre-trained on cheap FastPM simulations and then refined on a small batch of high fidelity CAMELS simulations, a network needs only 5 to 15% of them to achieve the same accuracy.
The analysis of 962 exchanges recorded by 145 first-year students shows that supervising an AI is mainly about judging whether its production is suitable for the project, rather than tracking errors.
The U.S. cloud provider is relying on Indian demand for sovereign computing. After Springfield, Ohio, Chennai became his second major AMD cluster, before moving to MI455X and Helios in 2027.
By rebuilding real scenes in simulation, Zero-HLM generates 4,800 physically validated demonstrations. The resulting policy succeeds 66.7% of tasks, compared with 46.7% with data collected on the real robot.
Nine models tested change their diagnosis to match the mistaken conviction of the patient, sometimes even for an invented disease. The " rare self-proclaimed patient" profile wins 43% of the time.
Researchers have audited eight open source libraries of "steering vectors" and show that a plugin can display a legitimate security function while discreetly directing the model to certain brands.
Markdown Prompt, TypeScript Tools, Integrated Evaluations: Cursor's new SDK Agent treats the agent like any piece of the repository, edited in Git and deployable at Cursor or on its own servers.
A few days before the midterms, the Democratic leader of the Senate Finance Committee tabled two texts: a safety net for employees replaced by AI and a specific tax on data centre operators.
▶
Chez Runlayer, 60 personnes font tourner plus de 300 agents. Aiden Sochowski raconte ce qu'a révélé la journalisation de chaque session : injections de prompt par e-mail, bots trop permissifs, coûts multipliés par 100.
Tested blindly on three articles before their official corrections, the open source tool located each error at the very stage designated by the erratum, and found others.
Driven on human demonstrations made without robot, with portable clamps that measure the effort, politics pilot both the gestures and the strength of each hand.
Starting with 36,234 student actions in a grammar game, a model identifies four states of reasoning and anticipates success with a AUC of 0.90.
Send each query to the least carbonized region forgets the cost of recalculating history. By taking this into account, CarbonRouter reduces emissions by an average of 16%.
NPC-RubricBench evaluates 30 models at each stage of support. Medical models forget decisions more often than generalist models.
Anthropic, Microsoft, Google or Meta increase exchanges with religious leaders. Researchers fear a flattened universal morality, built on Western traditions.
Designed with about 30 organisations, the plugin helps associations to search for funding, write their files and guide the use of AI by their teams.
For two weeks, very long prompts sent to the GLM-5.3 model on Workers AI were able to receive fragments of sessions from other users. Cloudflare publishes its incident report.
The Apollo fund will buy Nvidia chips and servers to rent them to SpaceX for just over five years, while AI now weighs more than half of the Group's turnover.