The paper teaches large language models to do what good students do: find where they went wrong, turn that lesson into a rule, and remember it for next time.
Search engines on social apps used to rely on many separate mini-models that often misunderstood slang and were hard to keep updated.
The paper shows a three-way no-win situation: an AI society cannot be closed off, keep learning forever, and stay perfectly safe for humans all at the same time.
Stroke3D lets you draw simple 2D stick-figure strokes plus a short text, and it builds a ready-to-animate 3D model with a skeleton and textures.
EcoGym is a new open test playground where AI agents run small businesses over many days to see if they can plan well for the long term.
ArcFlow is a new way to make text-to-image models draw great pictures in only 2 steps instead of 50, giving about a 40× speed boost.
This paper introduces GEBench, a new test to check if image generation models can act like real app screens that change when you click or type.
InternAgent-1.5 is a single AI system that can read papers, plan experiments, run code or lab steps, check results, and keep improving over time.
MOVA is an open-source AI that makes videos and sounds at the same time so mouths, actions, and noises match perfectly.
This paper teaches AI to write movie-like scripts for videos by adding exact timestamps and rich details about what you see and hear.
GISA is a new test (benchmark) that checks how well AI assistants can search the web like real people do.
This paper teaches language models not just to get the final answer right but to think in a way others can reliably follow.