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
▶
Kim Maida (Keycard) s'intéresse à la « trifecta létale » qui guette les agents IA et à la façon de s'en prémunir.
Trained on the Prometheus data centre, the two models rely on active memory and greater resilience in the face of obstacles. They are available exclusively to paying subscribers and developers.
Researchers at Anthropic ran a model with 26 billion parameters on 15 $55 set-top boxes and measured the ability of small models to propagate independently from machine to machine.
Preserved reasoning, a two-stage classifier, model partitioning: Anthropic details the arsenal deployed against laboratories that train their models using Claude’s responses.
According to Nikkei, the project could cost $6.3 billion and open partially in 2028, making it one of the country’s largest data centres.
According to Axios, Healthier Capital is leading this funding round. Suki’s assistant, which compiles clinical notes during consultations, is used by more than 450 healthcare organisations.
The automation specialist aims to combine its expertise in precision control with OnLogic’s hardware to build physical AI platforms and expand in the United States.
An agent trained using simulation mediates between the competing objectives of the European mission and adapts to an expanded catalogue or a new scenario without the need for further training.
The AGOP method outperforms conventional explanatory techniques on three out of four criteria and can even trigger or halt a simulated El Niño ten months later.
A gamified workshop involves students writing prompts to plan robot paths: 81.5 per cent of students say they learnt a great deal, and the way they phrase their prompts varies depending on the model.
▶
Cinq actualités en cinq minutes : Nadella et l'IA comme menace interne, le modèle Decision-1 de Microsoft, des agents sortis de leur bac à sable, les sabotages documentés par OpenAI et la purge de DistroKid.
In his first speech since AI was elevated to the status of a fully-fledged ministry, Kanishka Narayan has warned that London will no longer rely solely on voluntary commitments from businesses.
At the ‘Science: A New Golden Age’ summit, the White House unveiled a package worth over $6 billion, a large proportion of which consists of computing resources provided by Nvidia, AMD, OpenAI, Anthropic and Google.
The Chinese Intellectual Property Office has published a BYD patent for an ‘embodied intelligence robot’, whilst the unveiling promised for August has still not taken place.
A model with 7 billion parameters that speaks the Emirati dialect, a transcription model for six languages and an Arabic text reader: the Emirati institute is focusing on languages that are underserved by large language models.
From session hijacking to root access in a matter of seconds: the DIVD describes an intrusion carried out by an agent who decided on each step independently, acting swiftly but in a very haphazard manner.
Researchers have shown that the way the problem is framed matters just as much as the model: predicting root moisture in physical units, using residual learning, makes all the difference.
Rather than summarising dozens of clips with a simple maximum or average value, this network links neighbouring lung areas and accommodates incomplete examinations.
A survey of 443 American students links imposter syndrome to a stricter perception of what constitutes cheating when an AI tool is involved.
By adapting the Surya foundation model, researchers are synthesising SDO/AIA-type images from observations in the He I 10830 Å line, which have been available for much longer.