Papers159

This paper teaches a language-model agent to explore smarter by combining two ways of learning (on-policy and off-policy) with a simple, self-written memory.

Large multimodal models (LMMs) can look at pictures and read text, but they still miss tricky cases, like tiny chart labels or multi-step math.

GUI-Libra is a training recipe that helps computer-using AI agents both think carefully and click precisely on screens.

The paper shows that when training reasoning AIs with reinforcement learning, treating every wrong answer the same makes the AI overconfident in some bad paths and less diverse overall.

LongVideo-R1 is a smart video-watching agent that jumps to the right moments in long videos instead of scanning everything.

PyVision-RL teaches vision-language models to act like curious agents that think in multiple steps and use Python tools to inspect images and videos.

LLMs trained with simple rewards often latch onto just a few ways of solving problems and stop exploring, which hurts their ability to find other correct answers.

The paper builds a Computer-Using World Model (CUWM) that lets an AI “imagine” what a desktop app (like Word/Excel/PowerPoint) will look like after a click or keystroke—before doing it for real.

Fast weight models remember context with a tiny, fixed memory, but standard next-token training teaches them to think only one word ahead.

AI models that make CAD designs used to learn mostly from simple “draw-then-extrude” examples, so they struggled with real, complex parts.

Big idea: Make image-making AIs stop, think, check, and fix their own work so they get better at both creating pictures and understanding them.

TAROT teaches code-writing AI the way good teachers teach kids: start at the right level and raise the bar at the right time.