Updated daily708 verified stories
Research News tracker.
Material AI papers, studies, datasets, evaluations, and scientific advances with direct links to the underlying research.
- 30stories this week
- 708sources cited
- 708verified timeline entries
Verified sourcing
Every story links to the strongest available evidence: original sources when available, otherwise clearly attributed reporting.
Plain English
What happened, why it matters, and what to watch — without the jargon.
No filler
When the signal is thin, we publish nothing rather than padding the feed.
More stories
9 storiesInnovation
CEDAR proposes finite-state verification for language-guided embodied AI
A new arXiv paper describes CEDAR, a framework that converts natural-language constraints for embodied AI agents into executable finite-state representations.arxiv.orgInnovation
Study finds structured reasoning helps language models above a token threshold
An arXiv study reports that planning, checking and repair hurt a language model at very small output budgets but outperform a single-call system once budgets reach roughly 1,500 output-equivalent tokens.arxiv.orgInnovation
Preprint finds agentic AI performance depends on both task success and resource use
A new arXiv preprint comparing OpenClaw and NanoBot finds no statistically established winner on full task completion, but reports large differences in time and peak memory. The authors argue that agent evaluations should link outcomes to the resources and execution records that produced them.arxiv.orgInnovation
Preprint says generative AI broadened undergraduate research in a bioinformatics course
A three-semester qualitative study reports that generative AI helped undergraduate research teams develop distinct projects and coordinate across expertise, while students retained responsibility for checking and rejecting AI-generated contributions.arxiv.orgSecurity
LongGuard study finds safety guardrails lose more than half their unsafe-input recall on long context
An arXiv paper reports that safety guardrails’ ability to detect unsafe content falls sharply as input length grows, and proposes training-free mitigations that improve average results in its tests.arxiv.orgInnovation
WM-R1 trains mobile GUI agents inside world models
A new arXiv preprint describes WM-R1, a reinforcement-learning framework that trains mobile GUI agents entirely through world-model-generated interactions rather than repeated access to a real Android environment.arxiv.orgInnovation
HyQuant proposes keeping critical attention states in high precision to reduce LLM memory costs
A new arXiv preprint describes HyQuant, a hybrid-precision method that compresses most LLM attention data into low-bit formats while preserving selected tokens and local context in higher precision.arxiv.orgInnovation
Paper proposes localized reinforcement learning for AI-generated web applications
A new arXiv preprint proposes training AI coding systems with rubric-based feedback tied to specific code regions, reporting large benchmark gains for interactive web-app generation.arxiv.orgInnovation
SpikeOPD reports improved accuracy for sparse spiking language models
A new arXiv preprint proposes SpikeOPD, a distillation method for autoregressive spiking language models. The authors report average accuracy gains of 0.8, 1.7 and 2.9 points across three model scales while preserving the students’ sparse-compute profiles.arxiv.org
One useful briefing each week
Keep up with AI without living in the feed.
Get the week’s verified AI news, original data, useful tools, learning picks, and fresh AI jobs.
Reach people who are learning AI
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 jobSubmit an AI tool