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 CISPA study lists 15,300 validated injections on 11,700 web pages. Seven out of ten are hidden in metadata, headers or comments that are never displayed.
Published in Nature on September 16, the framework converts an article and its code into an MCP server, turning the publication into an executable tool rather than a fixed text.
A POLITICO/Public First poll conducted from September 13 to 15 also shows that a plurality of the population would prefer a pause in the development of advanced models.
BairesDev's quarterly survey measures 13 hours saved per week on code, but 67% of respondents now spend more time reviewing what the machine produces.
A test bench shows that the decision to yield the right-of-way varies according to the pedestrian's supposed gender, origin, age, disability, skin colour or social status.
In 11 days, largely autonomously, the model produced 13 million lines of Lean code and proved more than 30,000 intermediate theorems.
Adam Peruta (Syracuse University) argues that risk is not a visibly failed creation, but a creation polished enough to pass internal validation and then underperform.
Ricard Solé and his co-authors transpose a three-state epidemiological model to the spread of LLMs and describe a threshold beyond which dependence becomes collective and difficult to reverse.
One team replayed 35 LLM-based software engineering works: in 37 to 63 percent of cases, a single prompt on a next-generation model did better than the tooling released a year earlier.
624 episodes on 52 domains to assess whether a model can anticipate future research directions — with model rankings that thwart expectations.
Internal safety mechanisms initially survive innocuous fine-tuning, but LoRA-type protections eventually give way on a larger scale.
Three MIT researchers show that initially identical linguistic agents divide themselves into explorers, builders and coordinators, by simply modifying the setting.
Inherent, founded by four former Google DeepMind employees, relies on reinforcement learning rather than size to also assess the "taste" of a research agent.
StateMemBench submits 234 multi-session scenarios to existing memory systems. Scores remain low, and a simple state structure package provides most of the gain.
A benchmark submitted on August 20, 2026 puts 29 agent configurations against ten families of training algorithms, with four hours of GPU per job. Recursive self-improvement is still very far from the mark.
An analysis of the PubMed Central corpus detects markers of LLM use in 77% of articles in 2025, compared to 52% a year earlier — well above previous estimates.
Rather than inventing exercises, the authors extracted their tasks from the workflows of AI products already adopted by the market. The observation is stark.
The Reconstruction benchmark submits seven models to 643 papers from six disciplines, giving them only the list of references prior to publication.
The Mobius-v0 architecture stores knowledge in a shared memory and entrusts reasoning to iterative modules, with a gain of about four times in inference speed.
A test bench tested on 14 models shows that an accuracy of more than 95% does not protect against the influence of a plausible figure slipped into the question.