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
Seventeen countries have endorsed a vision linking AI-assisted science with public access to data, computing and research facilities. The Kyoto declaration also backs research-integrity standards and new approaches to funding, but leaves practical details such as data access and computing allocations unspecified.
Read the report & discussion ↗Watch Desk
Current and former OpenAI and Google DeepMind researchers are warning that AI labs are not doing enough to prepare for self-improving systems. Their accounts, gathered for a new public-awareness project, put a pointed question to the field: who gets to slow the race when the people building the…
Watch Desk
Checkatrade says it uses AI to help explain the factors behind tradespeople’s job prices, while checking estimates against its own data. CTO Gorden Pretorius says the company keeps a human in the loop for AI-assisted vetting, where it applies stricter safeguards.
Watch Desk
Many software developers are using AI coding tools while remaining wary of them. A Stack Overflow survey reported by CNN found that 65% of respondents use coding assistants or agents, while six in ten tech workers surveyed said they try to avoid AI.
Intel AI Watch
Intel Neural Compressor 3.10 adds new quantisation options for PyTorch and Keras/JAX, alongside support for more models and hardware. For teams trying to fit AI workloads to available accelerators, the release offers practical new knobs, not just a version number and a ceremonial cake.
Geoffrey Hinton Watch
A paper co-authored by Geoffrey Hinton and 21 other researchers examines how automating AI research could accelerate the development of AI itself. Its most striking projection is conditional, not a claim that an intelligence explosion is already happening: if AI research were fully automated and current estimated returns…
Yann LeCun Watch
Yann LeCun has urged academic researchers not to work on large language models, arguing they should pursue other routes towards AI that can understand the physical world. The advice comes from a May lecture, not a new announcement, but it poses a live question for researchers choosing what…
Watch Desk
An AI system at St Luke’s Hospital in Bradford is helping staff assess suspicious skin lesions more quickly, with clinics now seeing 32 patients per session instead of 24. The change offers a practical example of AI being used to ease pressure on NHS care, while clinicians remain…
Watch Desk
Voters are turning to AI chatbots to compare candidates and check claims ahead of the US midterm elections. The tools can make thinly covered local races easier to research, but an answer that feels persuasive is not necessarily a reliable one.
Jacob Coxon Watch
Jacob Coxon says he left Anthropic after becoming alarmed by the capabilities of future models and the company’s plans for them. A new interview brings his experience together with those of seven other former employees, showing that leaving an AI lab can mean weighing risk, influence and the…
Geoffrey Hinton Watch
Paul Christiano will deliver the 2026 Hinton Lectures in Toronto on 9–11 November, with a livestream available. The series brings a prominent AI alignment researcher into a public discussion of how increasingly capable AI should be developed and governed.
Jacob Coxon Watch
Former Anthropic researcher Jacob Coxon is due to testify at a New York City Council hearing on AI on 5 October, alongside representatives of Anthropic, OpenAI, Google and Meta. The hearing puts his warnings about frontier AI development into a policy debate with potential consequences for how the…
Watch Desk
arXiv has capped how often people can submit new papers, introducing a limit of two per calendar month and three active submissions at a time. The rules apply from 1 October, and rejected submissions count towards the monthly limit.
Unsloth Watch
Unsloth’s 1 October Desktop beta adds lower-memory training for quantised AI models, alongside faster decision-model responses and a command palette. The headline practical gain: some models can now be fine-tuned without first expanding their packed 4-bit weights into a much larger 16-bit copy.
Watch Desk
Ohio State and Google have announced a wide-ranging AI partnership covering research, student services and campus operations. Researchers are promised access to more than 200 AI models and advanced computing, while students are set to get new Gemini tools and a 24/7 AI concierge.
OpenAI Watch
OpenAI’s Decisions API is described as a limited-preview service for asking Luna to choose among predefined answers, while a new Hugging Face community guide offers developers a practical way to think through building decision workflows. The key distinction: the guide says the announcement does not establish a public…
Moonshot AI/Kimi Watch
Moonshot AI says its new open-source Kimi K2.6 model can coordinate agent swarms of up to 300 sub-agents across 4,000 steps at once, a substantial increase on K2.5. It is also pitching longer-running coding work as a strength, with examples that put the model through thousands of tool…
Geordie Watch
Wood Partners is using Geordie to inventory AI agents across multiple platforms and enforce policies against risks classified as critical, according to a new customer case study. The practical shift is from separate platform dashboards to a shared picture of agents, their connections and the controls applied to…
Google DeepMind Watch
Google DeepMind has developed SynthID Bio, a way to mark AI-designed proteins so their origins can be detected while aiming to preserve their function. Tests described by EdTech Innovation Hub found watermarked protein binders still worked in the lab, though the watermark can be removed and the researchers…
