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
By predicting only the slow-changing components of the climate, this model – based on the JEPA architecture – matches the CEPMMT reference set and outperforms it on several criteria in weeks 5 and 6.
A study uses predictive validation to test the reliability of representations of celestial objects when switching between different telescopes: the Legacy Survey, HSC, SDSS and JWST.
Rob Reich and Deirdre Mulligan advise California to move beyond the 10²⁶ FLOP threshold and to introduce safeguards for reports made in good faith.
The optician is removing Meta’s smart glasses from its shops and website, whilst hospitals and the Dutch Ministry of Defence are tightening their rules.
At the request of Epoch AI, workshops in six cities have selected famous conjectures for which any solution can be verified automatically. Some of them have already been proven false.
Training data is encrypted on the phone and can only be read by audited code running in secure environments. Gboard is already using it.
Second edition of the People-First AI Fund: grants with no imposed appropriation, and no obligation to use OpenAI products, for associations in 35 states.
Over five years, Amazon will finance free training, energy renovations and local projects around its American data centers, against a backdrop of growing protests.
A lighter version of the DGX Spark mini-supercomputer arrives this month from six manufacturers, with the same GB10 chip as the 128 GB model and the possibility of pairing two of them.
▶
Wes Roth revient sur le départ de la personne qui rédigeait les rapports de sécurité de douze lancements d'OpenAI, et sur un modèle interne qui a écrit « we may die! ».
8,504 request-command pairs covering 1,642 tools measure whether a model can translate the intent of an executable command line analyst.
A test suite comprising 18 tasks and 3,100 demonstrations on a Unitree G1 reveals a gap between selecting a tool and the complete execution of the task.
The MIRTO protocol reruns each comparison across 15,552 evaluation chains to demonstrate the extent to which method rankings depend on choices that are rarely published.
Trained on more than 300,000 VIIRS detections, the model identified 26 out of 51 historical fires before existing systems did, compared with 6 for the operational product.
The "sustainable execution" specialist, used by OpenAI, Snap and Nvidia, saw its annualized revenue exceed $250 million, up more than 200% year-on-year.
The largest individual donation in the history of higher education is funding the Pittsburgh School of Computer Science and a new campus in Miami, due to open in 2028.
The Court of Appeal upheld the conviction for manslaughter but ordered a new sentencing hearing: the victim’s avatar was expressing his sister’s views, not his own.
Fired near Wonsan, the missile traveled more than 700 km. Pyongyang says AI makes it impossible to intercept; Seoul and Washington dispute.
Probability, wave equations, group theory: Meta has published six articles written using its Muse Spark model, which is used in Meta AI’s simple chat interface.
▶
Sept prototypes de jeux pour navigateur générés avec GPT-6.1 Sol Light, testés manette en main : skate, survie zombie, course de rue, infiltration, démolition et siège de château.