Papers1055

This paper asks whether generation training benefits more from an encoder’s big-picture meaning (global semantics) or from how features are arranged across space (spatial structure).

Big AI models often write very long step-by-step solutions, but usual checkers either only check the final answer or get lost in the long steps.

This paper builds a math problem–solving agent, Intern-S1-MO, that thinks in multiple rounds and remembers proven mini-results called lemmas so it can solve very long, Olympiad-level problems.

Diffusion models sometimes copy training images too closely, which can be a privacy and copyright problem.

LEO-RobotAgent is a simple but powerful framework that lets a language model think, plan, and operate many kinds of robots using natural language.

This paper builds InternGeometry, a large language model agent that solves Olympiad-level geometry by talking to a math engine, remembering what worked, and trying smart new ideas.

T-pro 2.0 is an open Russian language model that can answer quickly or think step by step, so you can pick speed or accuracy when you need it.

Long texts make standard attention in large language models very slow because it checks every word against every other word.

This paper shows how to make home-helper robots better at long, multi-step chores by smart training on diverse tasks and by polishing the model after training using its own best attempts.

Robots often act like goldfish with short memories; HiF-VLA fixes this by letting them use motion to remember the past and predict the future.

UniUGP is a single system that learns to understand road scenes, explain its thinking, plan safe paths, and even imagine future video frames.

This paper introduces BiCo, a one-shot way to mix ideas from images and videos by tightly tying each visual idea to the exact words in a prompt.