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
The Super VL, with 120 billion parameters, is aimed at enterprise users, whilst the Bolt 74B-A9B and E4B are designed to run locally on RTX Spark and DGX Spark machines.
Projects such as dumbtv and cleanyourtv show that a code agent can now ‘clean up’ a Wi-Fi-connected television: adverts, tracking and unnecessary software are removed.
Several hundred thousand GPUs spread across several buildings, a network stretching over 100 km and a target of around one million GPUs: Meta offers a behind-the-scenes look at Prometheus.
The Chinese laboratory has announced scores of 73.7 per cent on DeepSWE and 82 per cent on CyberGym; the model, tested anonymously under the name ‘Space Bunny’, will publish its weights at a later date.
With the Rosalind Academic Plan, OpenAI’s model dedicated to the life sciences – which until now had been reserved for major companies such as Amgen and Moderna – is now becoming accessible to academic research teams.
The former Crescent Island has become the Intel GPU D-series: 1.3 petaflops in FP8, 1.5 TB/s of bandwidth and a massive amount of memory capable of handling models with 300 billion parameters on a single card.
The new version of Anthropic’s code agent generalises sub-agents that inherit the conversation context, speeds up the loading of plugins and adds a compaction setting in the event of inactivity.
In an interview with Joanna Stern, the Microsoft boss outlined a version of Windows in which each agent has its own identity, monitored by Defender, and a $2,600 RTX Spark laptop capable of running large models without relying on the cloud.
The Palo Alto-based start-up has unveiled a visual reasoning model that revisits its image representations whilst processing, charged at $0.60 per million tokens input.
Although it appears to be a minor update, the October revision of DeepSeek’s Flash model gains nearly 6 points on HLE and 11 on Terminal-Bench, whilst producing shorter responses.
The ‘ambient’ assistant listens in on conversations to learn its user’s habits and initiate actions without waiting for instructions. A dedicated connected device is planned.
Eight to ten key features have been deferred to PostgreSQL 20. According to committer Peter Eisentraut, LLMs have raised the bar for quality following the writing of the code, regardless of the coding ‘vibe’.
Microsoft’s new foundation model, an MoE with 80 billion active parameters, is said to be 4 to 12 times more efficient during training than DeepSeek-V4-Pro-Base. Its weights will be made available via Foundry.
Meta has released the weights of a mixture-of-experts model with 816 billion parameters, 42 billion of which are active, featuring a context of one million tokens and optimisation tailored for Apple chips.
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.
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.
Staff can now search for items, add them to a basket and pay via Shop Pay, but each purchase must be approved by the buyer, who can set a budget and conditions.
In response to games being decompiled by AI and then recompiled for the browser or Apple Silicon, Xbox has set up a small team tasked with openly experimenting with these techniques.
‘Komiya in the Interrogation Room’, which can be played for free in a web browser, was hailed by its creator as ‘truly astonishing’ but received a very lukewarm reception from some players.
A simple toggle will remove AI-generated tracks, albums and artists from search results and recommendations, whilst more than half of the daily uploads to the platform are now synthetic.