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
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…
Read the report & discussion ↗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…
NVIDIA Watch
NVIDIA is adding a 64GB unified-memory option to its DGX Spark AI workstation, with sales through partner manufacturers due to begin on 23 October. The useful twist is that two systems can be linked to pool 128GB of memory, giving local AI developers room to try larger workloads…
Anthropic Watch
Anthropic co-founder Christopher Olah reportedly urged Pope Leo XIV’s advisers to take the possibility of AI consciousness seriously before the Pope published a May address rejecting the idea. The episode, reported by The Telegraph on 2 October while citing The New York Times, puts a live question about…
Watch Desk
A Hugging Face community guide compares six reinforcement-learning methods used in language-model training, matching each to the problem it is intended to address. Its useful point: PPO, RLOO, GRPO, Pref-GRPO, GSPO and GDPO are not interchangeable recipe names; they make different trade-offs in compute, training stability and reward…
Watch Desk
Braze has launched several AI tools for marketing teams and signed a strategic collaboration agreement with Amazon Web Services. The package combines campaign decisioning, automated checks and conversational agents, with a new route for bringing Braze functions into other AI tools.
Watch Desk
A new class of AI models is being built to handle the quick, structured decisions around larger language models: routing requests, checking outputs and choosing what an agent should do next. The potential payoff is faster, cheaper routine checks, with a more capable model called in when the…
DeepSeek Watch
DeepSeek Harness version v0.2.1-alpha.1 is testing compatibility with Claude Code Mods, opening a possible route for extensions to work across AI coding tools. The bridge is experimental, not a plug-and-play way to move Claude’s mods over.
Watch Desk
Reddit plans to discontinue RSS feeds on 13 November and end public API access in March 2027, according to Decrypt. The changes could leave developers and researchers without familiar ways to access public Reddit data, while the company directs them towards its Developer Platform.
DeepSeek Watch
DeepSeek has released preview desktop apps for DeepSeek Harness, its open-source agent framework, with installers for Apple silicon Macs and 64-bit Windows PCs. The practical shift is that users can now try the harness through a desktop interface, with tools and plugin management built in, rather than assembling…
Watch Desk
Fine-tuning improved one narrow tool-use task in SCAI’s five-week experiment, but made a strong model worse at coding. The lab’s practical conclusion is refreshingly unspectacular: measure the problem, try cheaper fixes, and fine-tune only when you can show they have run out of road.
Watch Desk
AI could give more people the means to pursue ideas, but a flood of machine-generated discoveries may leave the world short of the institutions, experiments and infrastructure needed to act on them. That is the central tension in an essay by Hemanth Asirvatham and Elliott Mokski, who argue…
DeepSeek Watch
DeepSeek and Huawei have launched an open-source programming toolkit aimed at Huawei’s Ascend AI chips, offering developers an alternative software stack to Nvidia’s CUDA. The toolkit includes TileLang plus compute and communication libraries, according to Mobile World Live’s 30 September report.
Mistral AI Watch
Mistral says it is opening a Munich hub for Physics AI and Industrial AI, and plans to build one gigawatt of European compute capacity by 2030. The proposal links industrial AI research to a much larger European infrastructure ambition, though Mistral has not given a location, cost or…
Microsoft AI Watch
Microsoft’s ThinkingBox benchmark grades AI agents on the state they leave behind, then repeats each task 20 times to see whether success holds up. Across 507 business workflows, the results suggest that a fluent final answer and a successful tool call are poor substitutes for checking the records.
Google DeepMind Watch
Designer Ross Lovegrove used Google DeepMind’s Gemini in an experiment that generated hundreds of variations on a chair, but the chosen design proved heavy and not entirely ergonomic when made physically. The project is a useful glimpse of AI as a creative amplifier, not a button marked “finished…
Meta AI Watch
Meta’s Muse is instructed to build and update detailed profiles of the people in a user’s life, according to WIRED’s account of files extracted from the assistant. That makes the privacy question more specific than simply how much personal data an AI assistant can remember: it is also…
Google DeepMind Watch
Google DeepMind has released Gemini Robotics 2, a robotics model suite combining an embodied reasoning model with on-device controllers. Research lead Keerthana Gopalakrishnan says the work also tackles whole-body control, while the field still has a long way to go on getting systems to generalise across different robot…
