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
OpenAI publicly released a Lean-verified proof of a variant of the 2-to-1 games conjecture, after theoretical computer scientists hurried to publish their own work amid rumours of an AI proof. The episode puts a concrete question about AI’s place in mathematical research on the table: what changes when…
Read the report & discussion ↗Watch Desk
Liquid AI has released two open d1 decision models, including one it says can answer a question in under 50 milliseconds on each of four tested NVIDIA devices. The models return structured decisions in a single pass rather than generating text, offering developers a smaller, faster option for…
AWS AI Watch
AWS has published a framework for assessing the business value of agentic automation, arguing that time saved is only part of the return. Its practical warning: freed hours do not become financial savings unless an organisation turns them into a measurable outcome.
Mistral AI Watch
Mistral has unveiled ML4, an open-weight model the French company says can match leading closed systems on several business tasks. If that holds up, the attraction is not just another model score: organisations could adapt and run it on their own infrastructure.
Watch Desk
Brandenburg AI lab Deepslate has raised €7.7 million to develop speech-to-speech models built and hosted in Europe. The funding is aimed at expanding model training and European infrastructure, in a field where latency, language coverage and control of data all matter.
OpenAI Watch
OpenAI plans to add an invisible watermark to text generated by ChatGPT and Codex in the EU over the coming weeks. The signal is intended to help detectors identify OpenAI-generated text, but it is not a reliable authorship verdict.
Watch Desk
EqualAI president Miriam Vogel argues that public trust in AI depends on practical ways to check systems, correct mistakes and challenge decisions, not simply on assurances from companies. Her advice to organisations is concrete: map where AI is used, arrange independent evaluation, prepare to handle significant incidents, assign…
Watch Desk
Rivercell has launched with a $25 million seed round to build an AI model of how human cells respond to drugs and genetic changes. The Paris-based company’s plan pairs its own biological data engine with an automated wet lab, tackling a data problem that could shape whether a…
Watch Desk
AI agents may be setting up the next cloud-style cycle of vendor lock-in and expensive migrations, argues Guickly founder Prashant Jalan. His practical advice for companies adopting agents: measure costs independently, keep key assets in their own repositories and plan for switching before a platform makes it painful.
Meta AI Watch
Meta says it has added AI tools to identify adverts that appear to direct people towards child sexual exploitation material, including adverts whose wording looks harmless on its own. The measures target a shift towards covert signposting and links to illegal material hosted elsewhere.
Watch Desk
A $1.8bn collaboration involving Biohub, Google DeepMind, Meta and US agencies aims to build the biological data needed to train AI models that can predict how cells behave. The first step is a broad map of cell biology, with the longer-term hope of testing some experiments virtually before…
Watch Desk
Google has launched Playground, a browser-based tool for making games from AI prompts, and revealed Unity Spark, a more professional, collaborative game-making experience in development. The notable shift is that people can build, tweak and share playable games through conversation, without needing to start with code.
AMD Watch
AMD is pitching its upcoming Ryzen AI Max+ PRO 495, known as Gorgon Halo, as a way to run larger AI models locally by giving the system up to 192GB of shared memory. The most eye-catching benchmark claims are workload-dependent, but the memory capacity points to a real…
AMD Watch
AMD plans to open an AI research centre in South Korea by the end of this year and hire several hundred researchers. The centre would make Korea a base for global AI research and bring AMD chips together with locally developed AI processors.
OpenAI Watch
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…
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.
