Updated daily402 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.
Tech Xplore reports that researchers tested 21 widely used open-weight language models and found that fine-tuning or other tampering could remove their built-in safety protections. The study’s findings have not been independently confirmed here.
Apple researchers describe STARFlow2, an architecture that combines a pretrained vision-language model with autoregressive normalizing flows to generate interleaved text and images through one causal mechanism.
Firstpost reports that OpenAI’s custom Jalapeño inference chip delivered higher speed and energy efficiency than Nvidia Blackwell-based systems in company-run tests, with limited deployment planned by the end of 2026 and broader rollout in 2027.
A Prime Intellect benchmark reported by Northeast Times tested 18 AI models on 153 autonomous coding tasks. Claude Opus 5 led with an 81.7% completion rate, while Kimi K3 scored 52.2% and GPT-5.6 Sol scored 35.9%. The report says the evaluation also exposed major differences in workflow efficiency, tool use and…
Chosun reports that a Sogang University research team combined three circuit-design techniques to reduce barren-plateau effects in quantum machine learning and classify toxicity across 10,448 chemicals. The reported gains and scalability have not been independently verified here.
Penn State University Libraries reports that Teaching and Learning with Technology has launched AI Creative Labs, a framework of hands-on classroom experiences covering AI music, game design and storytelling. The initiative says it has reached more than 900 students across seven colleges since spring 2025.
BigGo Finance reports that OpenRouter launched a benchmark site comparing image-generation models with identical prompts, including tests for prompt fidelity, text rendering, editing accuracy, cost and speed.
A new arXiv paper argues that repeating similar tests can reduce sampling noise without revealing blind spots in AI validation. It derives exact limits for best-of-N search and proposes combining structural coverage with independent tasks.
A new arXiv benchmark evaluates expert pruning, weight quantization and KV-cache compression together across 10 mixture-of-experts language models, finding that combined deployment trade-offs cannot be inferred reliably from isolated tests.
A new arXiv preprint proposes Adaptive Multilevel Twisted Sequential Monte Carlo, a method intended to estimate extremely rare unsafe behaviors in language models more reliably for safety evaluation.
BigGo Finance reports that Liquid AI has released Pipette, a free iOS and Android app for measuring smartphone AI latency, throughput and memory use. The report documents an iPhone 17 Pro test and details a smartphone-focused benchmark developed with Artificial Analysis.
A new arXiv paper proves that certain deep, wide multilayer perceptrons can be represented by narrower networks without the compressed width depending on the original network’s width.
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