Artificial Intelligence News
AI research tests new methods for training, evaluating and understanding artificial intelligence. Find news on machine learning papers, reasoning, model behaviour and experiments that challenge existing assumptions. Read what researchers measured, how they tested it and where the limitations lie, with links to the original work and discussion of what the findings could mean beyond a benchmark.
Lead story
Ten years after its launch, the BBC micro:bit has reached an estimated 12 million devices, with about 73.5 million students learning with it across more than 85 countries, according to the Micro:bit Educational Foundation. Arm says the next chapter is using approachable, hands-on computing to help young people…
Read the report & discussion ↗vLLM Watch
vLLM has added bi-directional KV-cache transfer for multi-turn AI workloads, letting prefill servers reuse context held by decode servers instead of repeatedly computing it. In AWS’s benchmark, the change cut time to first token by up to 3.3 times for one tested model, though the gains depend on…
Watch Desk
Prime Intellect’s Prime Agent has been rewritten in Rust, adding native Windows support and on-disk session persistence alongside claimed gains in speed and memory use. The release is a substantial change for people building AI agents, though the performance figures come from the account of the release and…
Meta AI Watch
Meta says it is using otherwise idle capacity across its inference fleet to improve ad recommendations, while keeping the computing power reserved for ordinary service protected. The approach offers a practical way to do more AI work with hardware already in the system, rather than ordering more capacity.
AWS AI Watch
Amazon Web Services researchers say a technique called priming can turn pretrained Transformer weights into hybrid models that combine attention with state space layers, cutting the cost of pretraining those hybrids by over 99%. They also report that the resulting models matched the source Transformers’ reasoning benchmarks and…
Watch Desk
Three US Democratic senators have introduced a bill to create an AI Safety Board within the Department of Commerce. It would set up technical evaluations of AI systems that could threaten national or economic security, putting a concrete oversight proposal on the table as Congress debates how to…
Microsoft AI Watch
Microsoft researchers have introduced StudentSim, a system that simulates how individual learners respond to questions and tutoring. In tests across chess, English writing and maths, it outperformed the reported baselines, and a chess tutoring experiment suggests simulated student feedback could help improve tutors before they are tested with…
NVIDIA Watch
NVIDIA and Princeton’s Dao AI Lab have developed ReplaySSM, a method for speeding up inference in hybrid AI models by changing how their state is stored. NVIDIA reports up to 1.80× faster SSM decoding kernels at batch size 256, with larger gains for speculative decoding on tested 120B-class…
Meta AI Watch
Meta has launched Muse Wonder and Muse Wonder Full, adding two models aimed at demanding coding and agent tasks. Wonder is available through Meta’s Model API and Muse Code, while Wonder Full is exclusive to Muse Code subscribers.
Watch Desk
ASUS has unveiled a compact server built around NVIDIA Vera CPUs for agentic AI and reinforcement-learning workloads. The XA P2N-E2 is designed to handle the less glamorous, essential work around AI models: running code, using tools, sandboxing tasks and processing data.
Watch Desk
Anthropic researchers say they found a focused set of representations inside Claude that may help coordinate information for reasoning. A New Scientist feature describes why this has caught the attention of consciousness researchers, while making clear that evidence of information processing is not evidence that a chatbot has…
Alibaba Qwen Watch
Alibaba researchers have detailed QwenGyre, a framework for training AI agents on long software-engineering tasks. It dynamically shifts GPUs between generating agent runs and training, targeting expensive stretches when hardware would otherwise sit idle.
Watch Desk
The UK’s AI minister says a breach of British cyber defences by AI is “likely, if not inevitable”, and says government researchers are evaluating advanced systems including Google’s Gemini. His warning comes alongside plans to prepare for threats involving autonomous agents, though the report describes no specific new…
Marvell Technology Watch
Marvell has announced two CXL 3.2 memory expanders for AI infrastructure, adding six memory channels and support for sharing memory across as many as four hosts. The company says the new Structera X controllers can expand capacity for larger models and longer context windows, with sampling expected in…
Watch Desk
The UK government plans to use advanced AI to help defend public systems and is exploring stronger rules for identifying and reporting serious AI-related cyber incidents. AI minister Kanishka Narayan also says the government wants its security researchers to keep access to leading AI models before release, so…
Watch Desk
A University of Liverpool spin-out developing AI-powered materials discovery has raised £4m to advance its technology and commercial partnerships. The funding will help Matter Machine build its team and turn research into a platform aimed at speeding up the search for new materials.
Watch Desk
PacBio and TychoBio plan to use long-read RNA sequencing to build datasets from more than 10,000 samples, supporting AI research into RNA therapies for rare diseases. The collaboration aims to connect the design of candidate treatments with their effects across the transcriptome, including potential unintended effects.
Watch Desk
Researchers at Toronto’s University Health Network are using AI to build interactive 3D models of patients’ anatomy and tumours, aiming to help people understand planned procedures and surgeons prepare for them. The technology could make a difficult conversation more tangible, though it still needs clinical trials.
