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
Defra says it used AI to turn two days’ worth of workshop material into interactive online learning in less than a day. The trial shows how the department hopes to make staff expertise easier to share, while leaving subject-matter experts in charge of judgement and review.
Read the report & discussion ↗Watch Desk
Tata Consultancy Services says 80% of its future hires will have digital and AI skills, as AI-led work reached 10% of revenue in the September quarter. The figures offer a snapshot of how the IT services giant is reshaping its hiring and business around AI, though the company…
Watch Desk
An AI hiring agent at law firm Addleshaw Goddard ranked unusually polished CVs highly, while a working paper suggests candidates using the same AI model as a recruiter may be more likely to make the shortlist. The findings raise a practical concern for applicants and employers: AI-written applications…
Watch Desk
Researchers have built an AI model that identifies individual satellite components in radar images and tracks them across a sequence, even when a satellite rotates and a part disappears from view. The work, involving the Alan Turing Institute and the University of Birmingham, could help improve space monitoring…
Meta AI Watch
Meta has released an open-source SDK that connects its Muse AI agent to ESP32 boards and Linux devices such as Raspberry Pi computers. It gives developers a way to link Muse with screens, audio, sensors and other connected hardware, extending the agent beyond a phone app and into…
Watch Desk
The Trump administration has reversed a request that US AI developers keep new models from Britain’s AI Security Institute before release, Reuters reports, citing the Telegraph. If confirmed, the change gives developers more room to share models with UK safety testers before launch, a useful opening for international…
Watch Desk
Adtran has launched Mosaic One Agents, a suite of AI agents intended to investigate network problems, coordinate work and advance tasks towards resolution. The notable promise is that teams can set how much autonomy the agents have, from recommendations to actions within defined boundaries.
Cloudflare Watch
Cloudflare’s Clef-omni can score text, images, audio and video in a single decision call, according to MarkTechPost. The model could spare developers some transcription and captioning work, but the reported tests also show it is not a reasoning upgrade across the board.
Watch Desk
Several open AI models are using hybrid attention designs to reduce the memory and computing costs of generating text. A technical analysis compares eight approaches and reports a trade-off: lower costs and faster performance can come with weaker retrieval across long contexts and less deterministic generation.
Watch Desk
Meta’s Muse has a plush mascot called Jolly, and other AI products are giving their agents cuddly faces too. Business Insider’s 11 October feature argues that this cuteness can make AI feel more approachable, even as concerns about its effects on jobs and society remain.
Watch Desk
A cryptomining campaign has compromised more than 3,400 servers, including systems running AI services such as LiteLLM and Ollama, according to research from Lumen’s Black Lotus Labs. Operators should check exposure and patch affected software: the campaign was still infecting new victims when Lumen published its findings.
Nous/Hermes Watch
A Verge reporter is testing whether local AI can handle useful personal tasks without sending sensitive information to cloud services. Using Hermes Agent on an M5 Ultra Mac Studio, the reporter has sorted more than 400 Steam games and analysed financial records, while finding that setup and reliability…
Watch Desk
A Hugging Face community article describes three AI systems built around a shared, signed-address routing method: AlephLM, AlephLLM and the adapter library AMOE. Its most interesting claim is that the method can be used across language models and diffusion systems, though the results remain the article author’s account…
Watch Desk
A new essay by historian of science Pranav Sharma argues that today’s AI systems are being credited with understanding when they generate convincing language without regard for truth. His challenge is not just to the technology, but to what people mean when they call it intelligent.
Watch Desk
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…
