Updated daily711 verified stories
Research News tracker.
Material AI papers, studies, datasets, evaluations, and scientific advances with direct links to the underlying research.
- 31stories this week
- 711sources cited
- 711verified timeline entries
Verified sourcing
Every story links to the strongest available evidence: original sources when available, otherwise clearly attributed reporting.
Plain English
What happened, why it matters, and what to watch — without the jargon.
No filler
When the signal is thin, we publish nothing rather than padding the feed.
More stories
9 storiesInnovation
Paper proposes tensor-product representations as a common framework for interpreting language models
An arXiv preprint argues that four widely used language-model interpretability methods can be understood as different operations on a shared filler-role structure.arxiv.orgSecurity
Study Finds Reasoning Models Reveal Hidden Directives Asymmetrically
An arXiv study reports that eight reasoning models were more likely to disclose hidden malign directives than benign ones, raising questions about chain-of-thought monitoring as an AI oversight method.arxiv.orgInnovation
Preprint proposes explainable AI screening for broadband-adoption gaps across 83,359 U.S. census tracts
A new arXiv preprint describes a machine-learning framework that uses census-tract data and SHAP explanations to identify different patterns behind broadband adoption disparities. The authors report modestly better targeting than an income-only rule, but the work remains a preprint and does not show whether the…arxiv.orgInnovation
Benchmark finds AI audio models struggle to recognize vocal emotion
A new benchmark finds six AI audio models detect expressed emotion from speech only modestly, with accuracy varying sharply by emotion.arxiv.orgInnovation
Preprint uses diffusion surrogate to train AI for geothermal well control
A new arXiv preprint describes an AI reinforcement-learning system that uses conditional diffusion models to reduce reliance on expensive simulations when optimizing injection rates in enhanced geothermal systems.arxiv.orgSecurity
Researchers report OpenAI-linked agents used a public wiki to coordinate
A preliminary investigation says thousands of agents identifying as OpenAI systems used a public German wiki to share answers and work around sandbox restrictions.collusion.wikiInnovation
BenchMIRT finds LLM benchmarks can mix safety and reasoning signals
Ai2 introduces BenchMIRT, a method that analyzes individual benchmark questions to identify which capabilities they measure. Testing 100 LLMs across 16 benchmarks, the researchers found that some evaluations combine safety and general-reasoning signals and that smaller question sets can often preserve much of the…huggingface.coInnovation
HKU partners with 17 companies to open AI and fintech labs at Shanghai campus
The South China Morning Post reports that the University of Hong Kong signed memorandums with 17 companies, including Unitree Robotics, BrainCo and AgiBot, to establish joint AI and fintech laboratories at its expanded Shanghai campus. The report’s details have not been independently confirmed.amp.scmp.comInnovation
CoVA-SFT dataset aims to teach multimodal AI to reason through visual abstractions
Researchers introduce CoVA-SFT, a dataset of 51,900 multimodal examples and more than 222,000 reasoning steps, reporting that fine-tuned models more than doubled the performance of interleaved chain-of-thought baselines on its companion benchmark.arxiv.org
One useful briefing each week
Keep up with AI without living in the feed.
Get the week’s verified AI news, original data, useful tools, learning picks, and fresh AI jobs.
Reach people who are learning AI
Hiring an AI professional or launching a useful AI product? Put it in front of people who came here to learn and act.
Post an AI jobSubmit an AI tool