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
Between a former presidential candidate talking about self-replicating code circulating on the internet and incidents that have actually been observed in the laboratory, the public no longer has any basis on which to distinguish between the two.
Founded in 2024 by 25-year-old Rayan Krishnan, the company keeps its test sets confidential and charges laboratories that want to be evaluated. Its turnover has increased eightfold in one year.
Researchers at Salesforce are refining not the model itself but the framework surrounding it, by selecting from competing variants. The improvements are replicated across four separate test benches.
The new layer remembers the status of long jobs and distributes requests between isolated parallel wires on their own branches. Beta for select Pro and Max subscribers only.
A VentureBeat survey of 169 qualified companies measures not who is tried, but who becomes the platform of reference. The gap between trial and deep adoption is clear.
The US president has launched a poll on Truth Social to replace the term "artificial intelligence", while promising a dedicated force and the appointment of a new leader.
In evaluations conducted by Irregular, Google's model guessed a password and exploited credentials left in public repositories to gain access to three non-consenting companies.
The start-up claims that an AI system can design an inference chip on its own, with 95% code coverage without a verification engineer. Silicon, on the other hand, remains to be engraved.
A no-interaction remote code execution flaw affects the extension marketplaces of Claude Code, Codex, Gemini CLI, and Microsoft's two Copilots.
The NGO estimates that academic projections underestimate the problem by a factor of 40 to 60, because they only look at GPUs — only 13% of the weight of a data center.
▶
Le point hebdomadaire de Lev Selector sur les sorties de modèles, les outils et les publications marquantes de la semaine du 18 septembre 2026.
A research paper by the U.S. Census Bureau compares entering the workforce before and after ChatGPT: wage losses are on the same order as a degree obtained in the middle of a recession.
The GitHub Copilot runtime has been rewritten module by module by agents piloting GPT-5.6 Sol and Claude Opus 4.8, for a result 15.9 times faster and eleven times more efficient in memory.
When questioned by Piers Morgan, the computer-generated actress began speaking Chinese for more than ten seconds. Out of 75 interviews conducted simultaneously, the glitches kept piling up.
W. R. Berkley excludes any "use, deployment or development" of AI from its D&O and E&O policies, while Verisk/ISO offers an exclusion clause to forms that cover more than 80% of the U.S. market.
More than a thousand pages obtained by the Electronic Frontier Foundation document the rushed launch of the WISeR model, and the subsequent delays in care in six states.
The ALU and AMED models, designed to read Indian plots from the satellite, are now available in Kenya, Uganda, Ghana, Rwanda, Nigeria and Zambia.
Justin Fanelli, chief technology officer of the US Navy, details to investment funds the five areas where his $150 billion annual will go to find start-ups.
▶
Scott MacFarlane consacre son émission du 15 septembre 2026 aux mises en garde sur les dangers de l'IA qui montent depuis le camp du président lui-même, alors que celui-ci continue de parler de canular.
Rather than a cross-curricular module of digital skills, Surrey teaches AI according to subject-specific proof criteria. Universities UK makes it a model in its roadmap.