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
Jev AI’s new workflow guide shows how to use typed model decisions for tasks such as routing requests, checking safety and assessing answers, while leaving consequential actions to application code. Its useful rule is simple: let the model recommend a path, not quietly acquire permission to take it.
Read the report & discussion ↗Watch Desk
Investors appear to have priced in a 32.6% lasting productivity boost for software engineering from AI, according to a working paper by economists at UC Berkeley and the London School of Economics. That is a measure of market expectations, not proof that engineers are already getting 32.6% more…
Watch Desk
AI agents can give a factually correct answer and still mislead by crediting the wrong source. Multiverse Computing describes a new verification system that checks both whether a claim is supported and whether it came from the source the agent says it used.
Watch Desk
Anthropic and OpenAI are warning that advanced AI needs stronger safeguards, but an Associated Press investigation asks who might benefit from the way that warning is framed. Its account points to a live dispute over who sets the rules: governments, independent evaluators or the companies themselves.
Watch Desk
Current and former OpenAI and Google DeepMind researchers have told Reuters they fear AI companies are moving too quickly on systems that could improve themselves, while safety measures lag behind. Their accounts, gathered in videos for the nonprofit Palisade Research, put named researchers’ concerns and the project behind…
Watch Desk
TypeSafe AI’s Jev is being pitched as an alternative to chat-style AI, handling structured questions and decisions instead of simply generating text. Business Insider’s account of a San Francisco hackathon describes developers already trying it in coding, search and voice products, with speed and cost at the heart…
Watch Desk
A Berkeley Human Rights Center study argues that large language models can affect rights across law, journalism and education, and calls on developers and deployers to assess those effects. Its researchers interviewed 56 people in 24 countries, bringing evidence from professions and communities often left out of the…
AWS AI Watch
Vietnamese businesses are adopting AI, but a shortage of skills and organisational readiness could limit what they gain from it, AWS’s managing director for ASEAN told Vietnam News. His examples include a Vietnamese bank cutting a loan application from 150 fields to 20, and a dairy company using…
Geoffrey Hinton Watch
Geoffrey Hinton and Yoshua Bengio are among the authors of a new report urging governments to prepare for AI systems that could accelerate AI research itself. The authors say projects that would take people months could be automated by AI by 2028, while stressing that an intelligence explosion…
Watch Desk
NASA is using AI to help researchers find patterns in vast science datasets and to help mission teams make sense of operational information. The practical promise is less time hunting across scattered records, and more time putting them to work.
Watch Desk
OpenRouter is offering Jev Router, a tool that uses TypeSafe AI’s Jev model to judge the difficulty of a request and route it to a suitable AI model. GIGAZINE reports that it is intended to send simple tasks to cheaper models and reserve more capable, expensive ones for…
Watch Desk
Jev AI’s new support-ticket walkthrough shows how to use a decision model to propose a category, team and priority separately, while keeping consequential actions and uncertain cases under human control. Its examples are useful, but they are fictional demonstrations, not evidence of measured accuracy.
Anthropic Watch
Anthropic says Claude agents helped find a previously uncatalogued pattern around a known enzyme. The harder question is whether that amounts to a scientific discovery, and who gets to decide.
