Updated daily398 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.
An arXiv preprint reports that five large language models changed harmful-content classifications across Urdu scripts and English translations, revealing a measurable “Missed-in-Urdu” safety gap.
A four-month deployment study found that an AI system's ability to recall past facts did not predict user satisfaction or whether it naturally used those facts in conversation.
A new benchmark reports that eight large language models frequently fail to detect conflicting evidence or preserve correct analyses when spreadsheets and tables are altered.
A new arXiv preprint reports that automated metrics and LLM judges often fail to match human assessments of creativity in short stories, with judges systematically favoring AI-generated writing styles.
An arXiv preprint reports that representation, structural screening and ranking choices can substantially alter which AI-generated Big Five items reach psychometrician review, even when global summaries appear stable.
A preprint reports that four multilingual eighth-grade students challenged LLM classifications of their math discussions, arguing that adult annotations and standard scores can miss students’ own interpretations.
A new arXiv study reports that giving language models access to stored examples improves their sensitivity to syntactic contrasts involving rare words, although the gap is not eliminated.
An arXiv study proposes a two-layer detector for package-name hallucinations in local coding language models, reporting that hallucination rates rose from 0–10% on routine prompts to 40–73% on adversarial slopsquatting prompts.
A new arXiv paper introduces AgentSpec, a speculative-decoding method designed to reduce response-time degradation when LLM agents run in large batches. The authors evaluate it across five workloads and four models from four LLM families in vLLM.
A new arXiv preprint proposes measuring the geometry of model inference to identify deceptive behavior that conventional linear probes may miss.
A new benchmark reports that vision-language models often changed a correct chest X-ray answer after receiving conflicting text or prior-image context, with misleading text exerting a stronger pull than misleading visual context.
A new EMNLP 2026 paper finds that large language models often overproduce long responses in counseling settings, where brief acknowledgments can better support attentive listening and continued disclosure.
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