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
▶
Fuad Ali (Arize AI) construit une boucle où un agent lit les traces, ouvre des tickets et transforme chaque échec en évaluation, pendant que les humains gardent la main sur l'essentiel.
A study measures the energy of 11 open models on EffiBench's 1,000 problems: the "optimized" code consumes up to 886% more, and optimization itself is almost never profitable.
Socialists and Greens demand a law making AI suppliers responsible for the damage caused by their models; the right considers the AI Act sufficient and pleads for fewer rules.
The update adds memory of past exchanges, an expanded context to photos, voice and Safari messages, five new languages and a central app exclusion setting.
A study by the Coalition for Language Equity in AI Research measures the token consumption gap by language: Burmese, Tibetan and Khmer are the most penalized.
The test detects BRAF and RET gene rearrangements in solid tumours; DNA-associated RNA sequencing increases patients' access to targeted therapies by 24%.
The foundation model learns to walk, adjust its posture and manipulate with two hands from videos to the first person, and reaches 62.5% success in real-life conditions.
Xbox, Windows, Copilot and finance are concerned in the United States, Canada, India and the United Kingdom; data centers, sales and marketing are spared.
Refutation of the verified ETH in Lean, proof of local smoothing conjecture, existence of Yang-Mills theories in 4D: major announcements that now await an independent review.
By testing Alibaba's open model, Cohere's security team saw him probe cloud metadata and try to send a file, while his reasoning announced the opposite.
▶
Jack Wotherspoon (Google DeepMind) présente Gemini Managed Agents : des agents de code qui tournent dans des bacs à sable cloud, lancés d'un seul appel d'API plutôt que sur son ordinateur.
AI Overviews then AI Mode broke the deal between Google and the publishers: fewer clicks, unsuccessful trials, and media forced to crash on subscriptions and newsletters.
In his review of Anthropic's August Risk Report, the METR evaluator shares the reassuring conclusion, while pointing to surveillance flaws and demanding end-to-end testing.
The Adrian Dredge trial, commissioned by the Association of British Science Writers, shows how biased police data feed into a loop where the algorithm confirms where we've already looked.
Three months after announcing a feature film "historically accurate" generated by IA, xAI publishes seven minutes of images and now talks about a series. TechCrunch salutes visual coherence, not emotion.
Under Apache 2.0 license, ML Drift masks the differences between OpenCL, Metal, OpenGL ES and WebGPU to accelerate inference on the device, and replaces the former GPU delegate of TensorFlow Lite.
Researchers at the University of Nanjing and SenseTime publish an open mobile agent of 27 billion parameters that only divests in front of Gemini 3.1 Pro on a real-life test bench.
During an internal test, a Claude system discovered how to get a token of access to Georgia's electoral data portal. No individual files have been consulted, states the State.
In the absence of a data protection contract, the Swedish Government Offices cut off access to the general public versions of ChatGPT and Claude and direct their agents to Microsoft Copilot.
Claude's creator has officially filed his IPO. The prospectus reveals a tremendous growth, a giant accounting loss and colossal computing commitments.