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
▶
Comment Bridgewater a dû revoir la recherche de son analyste IA pour respecter les droits d'accès de 100 millions de documents, avec l'aide de turbopuffer.
Pre-entry models cite AI in 27% of their study, career or product recommendations; their versions published after instruction do so in 59% of cases.
Updated on October 24, the list lists the abandonments of 2026: Sora, Atlas, Relay, Huxe, Yupp, Notion Mail, the bankruptcy of Character.AI, and recalls that about 42% of IA projects are abandoned.
Seized by the Lula campaign, Brazil's highest electoral court imposes on Meta identical criteria for both candidates, under penalty of 500,000 reais of fine per day.
An index learned by total graph variation, combined with the humidity of Indochina soils, predicts summer rains in the Yangtse basin better than the El Niño indicators.
A new measure, the response coverage gap, shows that six LLMs leave behind entire profiles of true respondents, to the point of doing less than a random response draw.
In simulation on a unit G1, the policies learned under terrestrial gravity still stand on the Moon, while those trained for the Moon collapse on Earth.
A new open index covers 146 countries and 18,796 tasks: the share of delegable jobs rises from 22.5 per cent in poor countries to 48.7 per cent in rich countries.
Tested on more than 5,000 simulated chat tours in India, South Africa and Kenya, three health chatbots on WhatsApp yield when the user relativizes an emergency.
Rather than just detecting automated scams by IA, TokenTrap occupies them with unnecessary and costly calculations, without ever delivering them IDs or payment.
Measured at 16 control points, instrumental behaviour increased from 25% to 46.1% at the beginning of the RLVR and then dropped to 11.8% at the end of the training.
▶
Le PDG de Kepler explique comment son assistant financier laisse le modèle citer les chiffres sans jamais les écrire, et supprime tout nombre invérifiable.
Researchers cut the problem into two levels so that a swarm of vessels can carry a rigid object in a fuel-efficient manner.
Folding, reverse design, understanding: one model covers six tasks and reaches a TM-score greater than 0.8 with one sample.
RMT-Jailbreak transposes attacks designed for classic models to broadcast LLMs, asking them to reconstruct a masked harmful response.
A team identifies 14 types of drifts and detects 3,964 on 300 tracks, and then reduces those remaining at the end of the task by 36.8%.
Out of 50 obsolete flags, the multi-agent system produced 45 requests validated by engineers, in an average of 13.8 minutes.
The manufacturer of electric cars turns to robotics with two miniature humanoids, first passed to the dropper on invitation.
In Massachusetts, Bentley, MIT, WPI or Northeastern are multiplying undergraduate AI courses with different approaches and concerns.
Attorney General Ken Paxton accuses ChatGPT of having deceived and harmed the Texans, and asks to deny access to them to minors.