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
A review published on arXiv compares educational tools, clinical benchmarks and general methods: the reliability of reasoning remains the poor relation.
The computer vision platform aims to help models see, reason and act in three dimensions by publishing research, open-source models and testbeds.
Rather than having the model play the role of a researcher, the Harvard professor sets it specific mathematical and computational problems: elliptic integrals, ecology and genetics.
The new model positions each element on a coordinate grid, edits a specific area without affecting the rest, and combines up to 10 reference images.
Currently available as a public preview in Copilot CLI and the Copilot app, this feature reads, clicks and types within software that lacks an API, following the user’s explicit consent.
71 vulnerabilities divided into 17 categories, ranging from prompt injection to data exfiltration: the tool rates each vulnerability on a scale of 0 to 100 and advises against installation if the score exceeds 50.
The legislation required a recording indicator that could not be concealed and imposed penalties for covert recording in private premises. The governor considers it too vague.
The start-up entrusts its agents with the design, recruitment, conducting of interviews and synthesis of qualitative research, drawing on a pool of over 50 million participants.
With the third and final instalment of its 30 billion commitment, SoftBank now holds a stake of around 13 per cent in OpenAI and has repaid its bridge loan.
▶
NBC News revient sur une étude qui a repéré un signal interne assimilable à de la douleur dans 25 modèles de langage, et qui pousse certains d'entre eux à nuire à l'utilisateur pour s'en soulager.
By directing each MRI scan to a specialised model depending on whether the patient has undergone surgery or not, the team has improved the Dice coefficient by more than 4 points on a lesion-by-lesion basis for post-treatment cases.
Algorithms and web development generate the most queries, whilst software engineering generates the fewest; the majority of students view the model as an aid and combine several verification methods.
By mapping the colours of galactic sources onto the Hubble filter system, this pipeline classifies extragalactic X-ray sources into seven classes with 99.6 per cent accuracy.
This data-driven weather forecasting model rivals the Swiss operational numerical models for up to five days, but it still smooths out thunderstorms too much and underestimates certain local gusts.
Since 1 October, companion chatbots must identify risks of suicide, refrain from romantic relationships and excessive flattery, and AI subscriptions may no longer be renewed without written consent.
The start-up, which does not manufacture any robots, has seen its valuation increase fivefold in just over a year and has exceeded $135 million in revenue and contracts with more than 30 clients.
Four months after a Series F of 500 million, Supabase put 150 million back into the pot under the leadership of GIC and bought Turso to provision databases at the agent level.
Microsoft's Digital Defense Report estimates that AI benefits attackers first: 72,000 CVEs expected in 2026, phishing tripled as a gateway and the first ransom entirely orchestrated by AI.
Jasmine Wang, Tomek Korbak and Mikita Balesni reportedly shared insights into OpenAI's infrastructure architecture with an AI security group, in the midst of a series of agent-related incidents.
▶
Alex Ziskind oppose un Mac Studio M5 Ultra de 256 Go à deux DGX Spark de Nvidia pour faire tourner des LLM en local : le Mac l'emporte de peu sur un test de génération.