Search

CharacterFlywheel is a step‑by‑step loop that steadily improves chatty AI characters by learning from real conversations on Instagram, WhatsApp, and Messenger.

The paper introduces Rubric-ARM, a system that teaches two AI helpers—a rubric maker and a judge—to learn together using reinforcement learning so they can better decide which answers people would prefer.

Diffusion language models (dLLMs) generate several tokens at once but usually throw away lots of helpful clues each step—RCD keeps and reuses those clues.

This paper shows that giving an AI a safe, tiny virtual computer (a sandbox) lets it solve many kinds of problems better, not just coding ones.

Big all-in-one language models are powerful but too expensive to run everywhere, while small specialists are cheaper but narrow.

Olmo 3 is a family of fully-open AI language models (7B and 32B) where every step—from raw data to training code and checkpoints—is released.

SkyReels-V4 is a single, unified model that makes videos and matching sounds together, while also letting you fix or change parts of a video.

Robots often learn a bad habit called the vision shortcut: they guess the task just by looking, and ignore the words you tell them.

This paper shows a simple, repeatable way to teach general Vision-Language Models (VLMs) to understand e-commerce items much better without forgetting their general skills.

Real instructions often have logic like and first-then and if-else and this paper teaches models to notice and obey that logic.

KlingAvatar 2.0 is a system that makes long, sharp, lifelike talking-person videos that follow audio, images, and text instructions all at once.

VINO is a single AI model that can make and edit both images and videos by listening to text and looking at reference pictures and clips at the same time.

Kling-Omni is a single, unified model that can understand text, images, and videos together and then make or edit high-quality videos from those mixed instructions.

The paper fixes a hidden mistake many fast video generators were making when turning a "see-everything" model into a "see-past-only" model.

SkyReels-V3 is a single AI model that can make videos in three ways: from reference images, by extending an existing video, and by creating talking avatars from audio.

Unified Thinker separates “thinking” (planning) from “drawing” (image generation) so complex instructions get turned into clear, doable steps before any pixels are painted.

VIBE is a new test that checks how well image-editing AI models follow visual instructions like arrows, boxes, and sketches—not just text.

T2AV-Compass is a new, unified test to fairly grade AI systems that turn text into matching video and audio.

Image-to-Video models often keep the picture looking right but ignore parts of the text instructions.

Real life directions are often vague, so the paper creates a task where a robot can ask questions while it searches for a very specific object in a big house.