Updated daily397 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.
A revised arXiv paper proposes a learning framework for deciding which LLM APIs to query and which generated answer to deploy when API costs vary and only the selected answer’s downstream reward is observed.
An arXiv study combines two vision models with open-weight multimodal language models to resolve conflicting evidence in plant-disease images. The approach improved PlantDoc accuracy from 63.9% to 68.5% with Gemma, while tests on two Cornell robot-acquired datasets reached 98.9% and 99.3% accuracy. The researchers…
A revised arXiv paper proposes TSDS, a framework that stops local reasoning when an action stabilizes and sends uncertain actions to a cloud model. The authors report 43%-65% lower per-episode thinking compute on three of four tested tasks while maintaining stated guarantees on expected reward and cloud-call rate.
A Thai research team adapted Ai2’s open data-curation toolkit to create Mangosteen, a 47-billion-token corpus that the team says improved Thai-language model performance while removing much of the underlying web data.
Daily Excelsior reports that the UK and Ukraine agreed to give British researchers and companies access to battlefield data from Ukraine’s Avengers AI Labs, alongside pilot projects involving perimeter sensing and low-power chips.
A new arXiv preprint introduces CoDrift, a one-step generative policy framework that combines behavioral compatibility and value-seeking objectives for offline reinforcement learning. The authors report the best average rank across 73 OGBench and D4RL tasks in both offline and offline-to-online settings.
A new preprint studies agents that can transfer control to a stronger language model during reasoning, combining a Bayesian stopping rule with theoretical guarantees and a limited Qwen2.5-Coder validation.
A KDD 2026 workshop paper describes a deployed vision-language-model serving agent that handles unresolved content-quality cases, reporting higher review productivity, lower normalized operating cost and faster signal delivery.
A new arXiv paper proposes PuzzleKV, a training- and calibration-free method that independently compresses completed pages of large language model KV caches. It reports more than 96% of Full KV performance at about 60% of original storage, and more than 93% at 18.7% of original storage with quantization.
A new arXiv paper reports that a multi-armed bandit controller selected prompting strategies for LLM essay scoring with comparable accuracy to exhaustive search while reducing calls by 78.4%.
ChosunBiz reports that KAIST researcher Lee Sang-yeop described AI systems for predicting protein functions, identifying possible drug interactions and guiding microorganism-based production of biodegradable materials.
MIT researchers report a framework called CrysVCD that constrains AI-generated material formulas before atomic structures are created, improving reported stability rates and reducing computational screening costs for targeted crystalline materials.
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