Papers159

This paper fixes a common problem in reasoning AIs called Lazy Reasoning, where the model rambles instead of making a good plan.

This paper shows how to make text-to-video models create clearer, steadier, and more on-topic videos without using any human-labeled ratings.

The paper shows that a model that looks great after supervised fine-tuning (SFT) can actually do worse after the same reinforcement learning (RL) than a model that looked weaker at SFT time.

This paper teaches a model to make its own helpful hints (sub-questions) and then use those hints to learn better with reinforcement learning that checks answers automatically.

Large reasoning models got very good at thinking step-by-step, but that sometimes made them too eager to follow harmful instructions.

Large language models sometimes reach the right answer for the wrong reasons, which is risky and confusing.

TTCS is a way for a model to teach itself during the test by first making easier practice questions that are similar to the real hard question and then learning from them.

The paper teaches AI agents better by grading not just their final answers, but also how they think and use tools along the way.

The paper tackles a real problem: one-shot image or text searches often miss the right evidence (low hit-rate), especially in noisy, cluttered pictures.

The paper shows how to make AI think faster and smarter by planning in a hidden space instead of writing long step-by-step sentences.

MemOCR is a new way for AI to remember long histories by turning important notes into a picture with big, bold parts for key facts and tiny parts for details.

When training smart language models with RL that use right-or-wrong rewards, learning can stall on 'saturated' problems that the model almost always solves.