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
Common Sense Media says tests of ChatGPT’s teen experience found gaps in crisis referrals and parental alerts, prompting the group to urge OpenAI to keep teenagers off the service. OpenAI disputes the evaluation, and says some of the behaviour identified was intentional rather than a failure of its…
Read the report & discussion ↗Watch Desk
Ivo has launched Sage, an open model tailored to contract work, built with River AI on DeepSeek V4 Flash. The pitch is that specialist training can make a smaller, cheaper model useful for legal teams, though the performance figures come from Ivo, not an independent evaluation.
Anthropic Watch
A small video studio says Claude Opus 5.5’s advantage in making animated videos is its visual planning, rather than its ability to write the code that renders them. In the studio’s workflow, Opus designed the sequence and GPT 6.1 handled the coding, a division of labour that produced…
Watch Desk
Big tech firms have borrowed around $500 billion since January to fund AI chips, servers and data centres, as US government bond yields climb above 5.30%. Yahoo Finance reports that analysts see AI borrowing as one pressure on markets, alongside inflation and other forces.
Watch Desk
Researchers at a Federal Reserve Bank of New York conference presented findings that AI chatbots can make people more confident in their answers without making those answers more accurate. One experiment found participants followed the chatbot’s response 80% of the time, whether its reasoning was sound or not.
Watch Desk
AI safety cooperation between China and the US should go beyond reporting incidents after they happen, Bloomberg Opinion columnist Catherine Thorbecke argues. She makes the case for direct information-sharing between companies and researchers, as both countries push ahead with increasingly autonomous AI systems.
Google DeepMind Watch
Google DeepMind has announced Gemini 4 Argon, a model aimed at long software-engineering jobs and other demanding professional work. Its standout offer is the ability to generate up to one million output tokens in a single response, but access is still restricted and Google’s benchmark results remain company-run…
Powerful Medical Watch
Sant’Andrea University Hospital in Rome has embedded Powerful Medical’s AI ECG analysis into its emergency heart-attack pathway, automatically notifying cardiology when a recording is flagged as high risk. Powerful Medical says clinicians opened those reports in a median of under 17 seconds after the alert: the notable development…
Watch Desk
Monash University has joined Project Tapestry, a global AI collaboration focused on building models while members retain control of their data and computing infrastructure. The project says its Monash-BharatGen team has already trained a shared model across Australian and Indian sovereign computing systems, keeping training data within each…
Watch Desk
GitHub is rebuilding its Git infrastructure to handle the heavy, concurrent reads and writes associated with AI-driven software agents. Its internal benchmarks show up to 35 times higher write throughput, a potentially important change for teams running development work at agent scale.
Anthropic Watch
Anthropic’s reported $518 billion in computing commitments sharpens a practical question for businesses: are they ready to put AI systems to work, and to take responsibility when they go wrong? An Observer analysis argues that enterprise controls may lag behind the infrastructure and growth ambitions described in Anthropic’s…
OpenAI Watch
OpenAI says it is publishing a broad set of mathematical results produced by an internal frontier model, alongside material intended to help researchers examine and check the work. The release includes formalised proofs in Lean and details about how the results were obtained, making this more than a…
Meta AI Watch
Meta has opened Muse Gadgets, a project for developers to connect its personal AI agent to home-built hardware. The useful shift is from an assistant that lives in an app to one that could reach buttons, screens and sensors around the home.
Jacob Coxon Watch
Former OpenAI and Anthropic researcher Jacob Coxon told New York City lawmakers that frontier AI firms are pursuing automated AI research, warning that it could leave people with less control over what comes next. His testimony also put a concrete workplace concern on the record: he said most…
Watch Desk
Startup Conway Research has launched an invite-only beta of Underdog, an AI assistant that runs on users’ computers and plans to earn money from purchases rather than advertising. It pairs a privacy-first pitch with a distinctly practical question: what will users let an assistant do when it can…
Watch Desk
MIT Lincoln Laboratory researchers’ latest survey compares more than 120 commercial AI accelerators, up from 57 in its first paper. The work gives buyers and researchers a way to compare performance, power and different hardware designs in a fast-growing market.
