Large reasoning models can often find the right math answer in their “head” before finishing their written steps, but this works best in languages with lots of training data like English and Chinese.
This paper teaches AI to solve diagram-based math problems by copying how people think: first see (perception), then make sense of what you saw (internalization), and finally reason (solve the problem).
This paper shows a new way (called RISE) to find and control how AI models think without needing any human-made labels.
DataFlow is a building-block system that helps large language models get better data by unifying how we create, clean, check, and organize that data.
This paper teaches a vision-language model to first find objects in real 3D space (not just 2D pictures) and then reason about where things are.
Skyra is a detective-style AI that spots tiny visual mistakes (artifacts) in videos to tell if they are real or AI-generated, and it explains its decision with times and places in the video.
Nemotron-Math is a giant math dataset with 7.5 million step-by-step solutions created in three thinking styles and with or without Python help.
OpenDataArena (ODA) is a fair, open platform that measures how valuable different post‑training datasets are for large language models by holding everything else constant.
Reasoning tokens (the words a model writes before its final answer) help the model think better, but they are not a trustworthy diary of how it really thought.
DentalGPT is a special AI that looks at dental images and text together and explains what it sees like a junior dentist.
The paper shows that video AIs do not need long, human-like chains of thought to reason well.
VG-Refiner is a new way for AI to find the right object in a picture when given a description, even if helper tools make mistakes.