Updated daily401 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 new arXiv preprint describes SAFE-G, a multimodal question-answering framework that combines graph-based evidence retrieval with reinforcement learning to keep answers tied to supporting information. The authors report accuracy gains on two benchmarks, though the supplied abstract does not provide absolute scores…
An arXiv paper introduces TASSO, a method for adapting vision-language models to new tasks while reducing the loss of earlier abilities and zero-shot performance. The authors report improvements over existing methods in CLIP-based continual-learning benchmarks, but the source provides no numerical results.
A new arXiv preprint describes a model-based reinforcement-learning framework that assigns hospital requests to heterogeneous robots while adapting to forecast errors and newly arriving work. In a case study using real nursing-task requests from inpatient floors, the authors report near-complete service and lower…
An arXiv preprint introduces CIDER, a continual reinforcement-learning framework that, the authors report, lets one robot policy learn six household and industrial manipulation tasks while retaining previously learned behaviors.
A new arXiv preprint describes three deep-learning routes for classifying materials from multisensory touch data and reports that thermal cues were consistently important. The authors say direct classification approached near-perfect accuracy, while models designed to reflect human perceptual stages performed less…
A new preprint describes an influence-function method intended to identify which training samples shape predictions and estimate spectral errors when ground-truth data is unavailable during space-mission operations.
Researchers propose DELTA, an attention-based encoder that lets a quadruped robot focus on terrain evidence relevant to foot placement while keeping computation fixed as map resolution increases.
A new arXiv paper proposes samplers for discrete diffusion models whose required discretization steps depend on the target distribution’s internal dependence rather than directly on its dimension.
A new arXiv preprint introduces ConvergeFlow, a flow-based language model designed to end at valid token embeddings without a cross-entropy-trained decoder. The authors prove convergence under stated regularity conditions and report competitive results on OpenWebText, but the abstract does not provide benchmark…
A new arXiv preprint proposes training sparse local linear models around individual predictions while retaining much of a black-box AI system’s accuracy.
A new arXiv preprint presents a mathematical framework for analyzing how feature geometries align inside deep neural networks. Its experiments suggest that alignment depends on layer-specific transport, interactions and cancellation rather than following automatically from training or lower risk.
A new preprint proposes an ensemble-based approach that measures how strongly language models commit to answers, reporting competitive calibration and hallucination-detection results across three open models and four question-answering 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