Research
AI research made readable: new papers, benchmarks and findings from labs and universities, and what they could mean outside the lab.
Stories
-
Project ProtoStar aims to make publicly funded science easier for AI to use
Universities, research groups and Hugging Face are teaming up to release publicly funded datasets, models and code in formats that people and AI systems can readily find and reuse.
Source: Hugging Face
-
Google Research gives AI models a ‘sleep’ phase to learn without forgetting
Google researchers propose letting language models pause to sleep, turning short-term memories into lasting knowledge so they can keep learning without losing old skills.
Source: Google Research
-
Google’s chest X-ray AI for TB now screens patients in African clinics
Google Research says its tuberculosis X-ray models have become a certified clinical tool used in African clinics, while open versions help health ministries build their own.
Source: Google Research
-
Trimming a small model’s vocabulary cut one company’s AI costs by 77%
Dharma AI rebuilt the vocabulary of a small language model around what each token costs to serve, and its essay grader became cheaper and more accurate.
Source: Hugging Face
-
Google maps the world’s forest types to show what replaces lost forests
Google DeepMind and partners have released a free global map that sorts forests into natural, planted and farmed types, and tracks how they change year by year.
Source: Google DeepMind
-
Google AI estimates pregnancy stage from simple ultrasound sweeps like experts
A Google study found that health workers without sonography training, using cheap handheld probes and AI, could date pregnancies as accurately as expert sonographers.
Source: Google Research
-
Ai2’s LitPivot helps researchers reshape ideas as they read the literature
A new research tool, LitPivot, critiques a research idea against papers the researcher chooses and suggests revisions backed by excerpts from those papers.
Source: Hugging Face
-
Microsoft trains AI agents to finish database tasks reliably, not just once
Microsoft researchers turned a benchmark that checks what agents actually changed in a database into a training reward, making a model far more consistent.
Source: Hugging Face
-
Microsoft’s AI diagnosis tool beats doctors on past cases in Science study
Microsoft says its multi-agent AI system diagnosed medical cases far more accurately than doctors in tests published in Science, though it has not yet been tried with real patients.
Source: The Verge
-
NVIDIA’s Cosmos3-H-Dreams simulates real surgery with generative AI
NVIDIA released a research simulator that generates realistic video of gallbladder surgery in real time, responding as a user moves the surgical robot’s instruments.
Source: Hugging Face