From the archive
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
A randomised trial found that primary teachers using Oak National Academy’s Aila AI lesson assistant spent 24% less time preparing lessons, without a drop in the quality of their materials. The saving was about 49 minutes a week, though teachers used the tool for only a small share…
Read the report & discussion ↗Watch Desk
Aleph Alpha released Kolibri 1, a German-English open-weight language model with 78.1 billion parameters, on 3 October. Its Apache 2.0 licence allows users to modify and redistribute the weights, but running the full model still calls for substantial hardware.
Watch Desk
Cloudflare has released its first machine-learning models, Clef and Clef-flash, for tasks such as choosing from predefined options. TokenPost reports that both are available through Workers AI, with weights under the Apache 2.0 licence for download and self-hosting. The practical draw is speed and flexibility, though the hosted…
Watch Desk
AI makes it easier for organisations to produce fluent copy at speed. It may also make their messages more alike, and the evidence cited by a Forbes Councils contributor suggests that this sameness can show up even when individual work improves.
AMD Watch
AI agents may be shifting some of the computing workload back towards CPUs, giving AMD and Intel a larger role alongside the GPUs that have dominated the AI boom. The change matters because agents need processors to keep their virtual computers and tasks running, not just chips to…
Watch Desk
Aragón is courting more than €60 billion in data-centre and digital-infrastructure investment over the next decade, with fast-track permits helping make the Spanish region a major draw. The speed could help build infrastructure for cloud and AI, but residents and researchers question whether public scrutiny can keep pace.
Watch Desk
Workers who use AI regularly often say it helps them work faster and more creatively, but the benefits are not evenly spread across the workforce. A US study found that workers with degrees and managers were more likely to use AI regularly, pointing to an access gap alongside…
Watch Desk
A new book by technology journalist Kevin Roose examines how rivalry, safety concerns and genuine excitement coexist inside leading AI companies. The New York Post’s account offers a vivid, if sometimes startling, look at the people shaping a race whose consequences reach well beyond the labs.
Mistral AI Watch
Morocco’s digital transition ministry and Mistral AI have developed two AI tools aimed at helping software understand Moroccan Arabic: a dialect identifier and a speech-recognition model that can transcribe conversations switching between Arabic, French and English. The planned open-source release could give developers a useful building block for…
Watch Desk
Mearns Castle High School in East Renfrewshire has appointed Jamie Laycock to a dedicated AI literacy post, with lessons for pupils and training for colleagues. The approach is less “AI is the future” than “check what it tells you”, which is a more useful starting point for a…
Stanford HAI Watch
Stanford HAI’s 2026 AI Index finds AI capability and adoption advancing quickly, while responsible-AI reporting remains patchy and documented incidents have risen. The report gives readers a broad, data-rich view of where the field is moving, and where measurement is still playing catch-up.
Google DeepMind Watch
Google DeepMind staffer Paige Bailey demonstrated Gemma 4 running locally in a web browser, with no API call or data sent off the device. The talk also laid out a range of model sizes and on-device tools, putting the practical case for open, smaller AI models on display.
Microsoft AI Watch
Bristol Myers Squibb says AI now influences all its small-molecule drug designs and has helped improve manufacturing yields by up to 40% for certain medicines. The pharmaceutical company also says it has 25,000 AI agents in use across the business, offering a substantial, if company-reported, picture of AI…
AWS AI Watch
Amazon has made Nova 2.5 Sonic generally available, with AWS touting better reasoning, instruction-following and tool-calling for real-time voice agents. Developers can use it in four AWS Regions, at the same price as Nova 2 Sonic, and pair it with Strands Bidi Agents, now generally available.
Berkeley BAIR Watch
A University of California, Berkeley team has built an open-access database of millions of local ordinances across all 50 US states, using AI to turn a sprawling archive of legal documents into searchable data. The work could help researchers and the public compare rules that are usually scattered…
Aleph Alpha Watch
Aleph Alpha has detailed Kolibri’s design for German-speaking public administration and industry, including document-grounded answers, native tool calling and adjustable reasoning. The practical proposition is a model built for infrastructure customers control, rather than another assistant whose operating arrangements arrive as a surprise. Its 5 October announcement adds…
