UniReason is a single, unified model that plans with world knowledge before making an image and then edits its own result to fix mistakes, like a student drafting and revising an essay.
SLIME is a new way to train chatbots so they follow human preferences without forgetting how to write well.
The paper introduces UnifiedReward-Flex, a reward model that judges images and videos the way a thoughtful human would—by flexibly changing what it checks based on the prompt and the visual evidence.
SWE-Universe is a factory-like system that turns real GitHub pull requests into safe, repeatable coding practice worlds with automatic checkers.
The paper shows that three popular ways to control language models—fine-tuning a few weights, LoRA, and activation steering—are actually the same kind of action: a dynamic weight update driven by a control knob.
This paper proposes ReSID, a new way to turn items into short token codes (Semantic IDs) that are much easier for a recommender to predict.
LatentMorph teaches an image-making AI to quietly think in its head while it draws, instead of stopping to write out its thoughts in words.
The paper fixes a hidden mistake many fast video generators were making when turning a "see-everything" model into a "see-past-only" model.
This paper studies how AI agents get better while they are working, not just whether they finish the job.
The paper introduces VDR-Bench, a new test with 2,000 carefully built questions that truly require both seeing (images) and reading (web text) to find answers.
This paper fixes a common problem in reasoning AIs called Lazy Reasoning, where the model rambles instead of making a good plan.
Loop-ViT is a vision model that thinks in loops, so it can take more steps on hard puzzles and stop early on easy ones.