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.
A study of more than 2,000 language-model training runs says scarce target data can be repeated 15–20 times in mixtures, with the best rate varying by scale and compute.
Apple researchers describe LINK, a pretraining intervention that replaces selected English words with word-level translations from a target language. The paper reports improvements across eight languages and five model sizes, including up to a twofold speedup in reaching equivalent downstream performance.
A position paper reports tacit collusion by DeepSeek-R1 agents in a simulated Bertrand pricing market, even after human prompts against collusion. It argues that observed-behavior certification should precede deployment of reasoning agents in economic markets; the evidence and safeguards remain preliminary.
A systematic review surveys how large language models are being studied for mental-health analysis, risk assessment, therapy support and multimodal monitoring, while stressing unresolved ethical and regulatory challenges.
An ICML 2026 position paper analyzing 500 Hugging Face model cards argues that open-weight foundation models need coordinated model cards, acceptable-use policies, and licenses to address safety and governance gaps.
A position paper proposes the Transition Complexity Profile, a standardized way to describe how unpredictable and long-range the dynamics of game environments are for game-world modeling and reinforcement learning.
A new arXiv benchmark places 15 language-model agents in a 20-year football-management simulation, testing whether they can make consistent decisions when short-term choices affect long-term outcomes.
A new arXiv benchmark reports that changing only the retrieval method raised a fixed model’s exact accuracy on financial reconciliation cases from 2.05% to 72.44%. The study also finds that a correct root-cause label often does not mean the system returned sufficient evidence for an auditable diagnosis.
A new arXiv study presents scaling-law experiments for text-to-image diffusion models across compute budgets from 10^19 to 10^22 FLOPs. Its authors report that image models need substantially more data per parameter than language models to train efficiently.
A new arXiv paper describes a self-supervised EEG method that reports 92.4% accuracy distinguishing Alzheimer’s disease from cognitively normal controls on the ADFTD cohort.
An arXiv paper introduces Agentic ESOpt, a proposed evolution-strategy framework for fine-tuning long-horizon language-model agents with inference-level GPU memory. The authors report gains for Qwen-3.5-27B on WebArena-Lite and improvements in 28 of 36 prompt-optimization settings.
An arXiv paper argues that mean squared error can misjudge irregular time-series forecasts and proposes a continuous-time metric tested across synthetic, semi-synthetic and eight real-world datasets.
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