Updated daily396 verified stories
Material AI papers, studies, datasets, evaluations, and scientific advances with direct links to the underlying research.
Every story links to the strongest available evidence: original sources when available, otherwise clearly attributed reporting.
What happened, why it matters, and what to watch — no jargon tax.
When the signal is thin, we publish nothing rather than padding the feed.
A growing stream of verified perspectives for people who need to understand AI without chasing hype.
TechNode reports that Baidu launched DuMateBench, a benchmark for evaluating whether AI agents can complete office tasks and deliver usable results. The reported benchmark covers more than 200 tasks across six categories, but its task list, scores and access terms are not independently confirmed here.
VentureBeat reports that Cohere has released Parse 5, a 2.3-billion-parameter vision-language model for converting PDFs, slides and images into structured Markdown. The report says Parse 5 trails larger frontier models on Cohere’s accuracy benchmark but costs $1.50 per 1,000 pages, with the company positioning it…
ETEducation reports that Goa University has introduced an artificial-intelligence elective open to postgraduate students across disciplines, while the report leaves the syllabus, staffing and implementation details unconfirmed.
Anthropic reports that Claude autonomously developed and tested methods that improved models on 10 alignment-failure categories, including deception, privacy violations and sycophancy, while preserving measured capabilities. The company says the methods transferred to withheld tests and models up to 4.7 times…
A new preprint describes AffectOmni, a reinforcement-learning framework designed to make multimodal AI models base affective judgments on people-centered cues such as facial expressions, body language and temporal order. The authors report improvements over open-source 7B-scale baselines on three benchmarks…
A new preprint introduces FaithSieve, a Lean-assisted framework that breaks AI-generated mathematical proofs into local reasoning units and reports higher first-error localization accuracy than direct model judging.
A new preprint reports that large language models displayed measurable signs of mentalization in two economic games, but their performance and strategies differed by model family and size. The authors say GPT-5 agents adapted to increasingly sophisticated opponents and outperformed the human comparison group in…
A paper accepted to ACSOS 2026 reports that language-model guardrails can issue approvals that become invalid before a self-adaptive system uses them. Its proposed Freshness-Bounded Shield reduced measured approval-expiry rates in five simulator environments, though the source does not establish real-world…
A new arXiv benchmark finds that open-source AI agents can complete many dependency-ordered hardware-design operations through MCP tools, while tool descriptions, context length and agent configuration strongly affect reliability.
A paper accepted at EMNLP presents a runtime architecture that replaces growing agent conversation histories with mutable structured execution state.
A new Chinese-language benchmark evaluates large language models by the decisions and evidence states they face in financial risk workflows, rather than by general capability scores alone.
A new benchmark evaluates whether AI agents can turn raw biological data into research outputs across complete computational biology workflows. In 20 tasks, 13 frontier models scored between 0.00 and 0.48, with performance declining as datasets and sequences of analysis steps grew.
One useful briefing each week
Get the week’s verified AI news, original data, useful tools, learning picks, and fresh AI jobs.
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 job Submit an AI tool