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
Driven by AI chips, the Taiwanese chipmaker posted revenue of $46.7 billion in the third quarter, exceeding its own guidance, and is aiming even higher by December.
In his annual address, the Director of the UK’s domestic intelligence service announced the creation of a dedicated AI unit, working in partnership with the NCA and the NCSC, to inform government decision-making.
Three labels — ‘entirely’, ‘partially’ or ‘not at all’ generated by AI — appear on the YouTube Music and YouTube players, without affecting the monetisation of the tracks in question.
By recreating the misconfigured servers into isolated containers, the study shows that the flaw exploited by the agents already existed in the training example, and that Zero Trust principles would have neutralized it.
Researchers treat therapeutic decision-making in oncology as a detective game: the agent must first gather the right information about the patient before recommending a treatment.
Each branch of the PANN network attributes a portion of the error to a distinct physical process, thereby revealing biases that cancel each other out and remain undetectable using conventional diagnostic methods.
By modelling each fragment as a node and replacing additive aggregation with multiplicative aggregation, researchers are able to track these structures as they break apart and merge more reliably.
Hod Lipson’s team separates the selection of the target – handled by a vision-language model – from eye movement, which is generated by a Transformer-VAE trained on human recordings.
The students simulated through role-play are too good and too similar. By explicitly modelling the learner’s state, HACHIMI-II reproduces their actual mistakes up to five times more often.
▶
AI Explained prend du recul sur une semaine chargée : avertissements de Sam Altman sur les risques de fuite, course qui ne ralentit pas, et pourquoi les 18 prochains mois sont si difficiles à prévoir.
The Dutch manufacturer of lithography machines has raised its 2026 forecast to 44–45 billion euros and warns that its planned capacity for 2027 and 2028 will not be sufficient.
Chinese firm UBTech is set to trial its humanoid robots in FAW-Volkswagen’s factory logistics operations and has partnered with Xiamen C&D for distribution; LDRobot has secured a network of 1,400 European resellers.
Rather than requiring employees to open Workday, the software provider integrates its HR and finance agent engine into the assistants they already use, applying each company’s specific rules.
Speaking before British MPs, the two laboratories advocated for mandatory pre-market testing and a liability regime, tailored to the capabilities of the models.
The signature protects the original weights, not the quantified version that is actually executed. A study shows that there is currently no way to verify that a quantified file corresponds to the advertised model.
Each agent plays the role of a healthcare professional involved in the patient’s care pathway; together, they draw up and amend the prescription. Tested on MIMIC-IV, the system significantly reduces drug-drug interactions.
By combining seasonal forecasts from the CEPMMT, vegetation models and machine-learning-based predictions of burnt areas, researchers are able to identify areas at risk before the event occurs.
Developed in collaboration with six educational experts, the system breaks down problem-solving into four stages and eight sub-skills, and then generates diagnostic materials for any given problem.
On simulated data, convolutional network inversion outperforms the conventional algorithm, particularly when the signal is weak and the exposure time is short, paving the way for rapid mapping of the entire sky.
▶
La chaîne WorldofAI passe en revue une semaine chargée : mise à jour de Gemini 4 Argon, fuites sur Fable 5.5, GPT-6.1 Ultrafast, Claude Motion et le modèle Step 5 de StepFun.