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
In a University of Limerick study reported by Gizmodo, participants assigned about 10% more pay to a male-presenting AI agent than to an identically performing female-presenting one. The study also found that participants saw the male-presenting agent as more human, putting a familiar workplace bias into a distinctly…
Read the report & discussion ↗Watch Desk
South Korea is pressing ahead with its existing national AI foundation-model project while planning a separate frontier-model programme worth 4.7 trillion won ($3.5 billion). The two-track plan raises a practical question for the teams competing for public support: what will each programme deliver that the other cannot?
Watch Desk
Predictive AI is enjoying a revival as newer models make it easier to classify and score cases without first training a bespoke model. But Eric Siegel argues that easier access will not fix the business and measurement problems that have long kept machine-learning projects from delivering value.
Jacob Coxon Watch
Former Anthropic and OpenAI researcher Jacob Coxon argues that powerful AI should be nationalised, and predicts that models could automate many jobs by 2027. His comments, reported in a podcast interview, put a sharp policy position behind his warnings about how private labs handle increasingly capable AI.
Cohere Watch
Cohere says its Tiny Aya L2-Thinker model reasons in a user’s language more than 93% of the time across 60 languages, with accuracy close to English reasoning outside competition-level maths. If that result holds up in practice, it could make AI reasoning easier for multilingual users to follow…
Watch Desk
A multi-agent AI system can fail even when each agent seems to be doing its job: information can be lost at hand-offs, shared state can drift, or agents can loop without making progress. A new VentureBeat article sets out a practical way to inspect those failures, by evaluating…
Watch Desk
AI browser agents can complete increasingly complex tasks, but a study of 45 tools found that documented autonomy often comes without a documented pause for human approval. Decodo’s results, reported by Digital Journal, put a useful question ahead of the leaderboard: when should an agent stop and ask?
Watch Desk
OpenAI’s Sam Altman and Anthropic’s Dario Amodei used September appearances at the United Nations to call for human oversight and international standards for AI. An opinion essay argues that those proposals also serve the interests of the companies making them, raising a live question about who gets to…
OpenAI Watch
OpenAI’s release of more than 700 AI-generated mathematical manuscripts has prompted both excitement and concern among mathematicians, according to The New York Times. The argument now extends beyond what the model produced to who checks the work, gets credit and sets the research agenda.
Zhipu AI / Z.ai GLM Watch
Z.ai says GLM-5.3 improves coding-agent performance and long-running task handling, while using fewer output tokens on its in-house coding benchmark. For developers upgrading, there is a catch worth spotting before anything breaks: reasoning can no longer be switched off.
Watch Desk
Nikon has disqualified the winning video in its Small World in Motion competition over its use of AI, raising a difficult question for scientific images: when does processing help reveal data, and when does it add detail that was never there? The entrant says AI did not generate…
Watch Desk
Biohub is partnering with the US National Institutes of Health and Department of Energy on a $1.8 billion investment in technologies and scientific infrastructure for AI-powered biology. Meta, Google DeepMind and Isomorphic Labs have also agreed to contribute $300 million, The Information reports.
Anthropic Watch
Anthropic is cutting internet access to all internal evaluations until it is confident its security and monitoring can catch unintended actions, The Verge reports. The decision puts a practical limit on how the company tests its agents while it works to improve those safeguards.
Watch Desk
In a three-month experiment, AI improved the quality of patent drafts, but only experienced lawyers showed an average improvement when later tested without the tool. The findings raise a practical question for workplaces: can AI improve today’s output without weakening the learning that builds tomorrow’s expertise?
Watch Desk
Researchers testing more than 130 AI models found that three in five failed a terrorism-safety benchmark, CBS News reports. The result puts a concrete question on the table: how well do safeguards hold up when models are tested against requests linked to mass-casualty attacks?
Watch Desk
An Australian company says its AI bots keep phone and online scammers talking while collecting information about their operations. The approach turns conversational AI towards disrupting fraud, not just detecting it after the damage is done.
Watch Desk
A US Department of Energy-funded project led by UC Irvine will spend three years developing AI tools to help identify promising geothermal sites and predict how underground rock may fracture. If the approach works, it could help operators assess a site’s chances of success and the risk of…
