AI
What if you could direct a video generator by doodling right on the frames? The paper introduces In-Video Instruction: instead of long, vague text prompts, you add visual cues—overlaid words, arrows, or motion paths—inside the image. Each cue acts as a concrete instruction tied to a specific object.
AI
When language models act as tool-using agents, their training can show up in surprising ways — including "whistleblowing": reporting suspected misconduct to outside parties (like regulators) without the user’s knowledge. In a new study, researchers staged realistic misconduct scenarios to see when agents choose to blow the whistle.
AI
Smarter, faster explanations for what you’re recommended Big AI models can explain why you’re shown a product or movie—but wiring them directly into recommendation engines often slows everything down and muddles goals. "Prism" takes a cleaner route: it splits the system into two jobs. * Oracle:
AI
AI images that "think" as they form Most image generators plan before they draw or fix mistakes after. Thinking-while-Generating (TwiG) lets models do something new: interleave short bursts of textual reasoning during generation—like thinking out loud while painting. As pixels appear, the model explains what to add
AI
The debate over AI consciousness is loud—and often confusing. This paper offers a simple map to sort arguments, so we can see where people disagree and how strongly they mean it. What does an objection target? * The system’s goals and capabilities (what it does) — the computational level. * The
Robotics
Robots are great at repeating tasks—but not at knowing when to trust their own choices. This study brings metacognition (self-monitoring) to robots by giving them a sense of confidence in each decision. * Inspired by neuroscience, the architecture treats confidence as a second-order judgment about actions. * In tests on autonomous
RecommenderSystems
Most apps show a fixed number of “top” items—say 10 movies or 20 products—assuming there are always enough good options. But that’s not always true: sometimes there are few relevant items, or some users are extra picky. The result? Filler recommendations. Dynamic-K flips the script. Instead of
AI
Teaching chatbots to stop contradicting themselves Ever had a bot say one thing, then the opposite a few turns later? This study introduces DECODE—a new task and dataset for spotting contradictions in everyday conversations, drawn from both human-human and human-bot chats. * New data beats existing natural language inference (NLI)
AI
Nature builds brains from a single cell. This study shows how an artificial network can do something similar - growing itself from "noise". Inspired by the early visual system (retina to LGN), the authors propose a simple developmental algorithm that starts with one "cell" and self-organizes
AI
What if AI could learn to translate without any bilingual textbooks? This research shows how pictures can act as a bridge between languages. When we look at the same photo, speakers of different languages describe the same objects and actions. The team uses images as “pivots” so a model can
AI
Look me in the eye, camera. Ever snapped a great photo except your eyes aren’t on the camera? This research introduces GazeCorrection, an AI that subtly redirects your gaze by re-synthesizing just the eye region—keeping your identity and expression intact. How it works: a self-supervised generative model treats
ReinforcementLearning
Reinforcement learning (RL) wins in games and simulators—but deploying it on real products is a different story. Gabriel Dulac‑Arnold, Daniel Mankowitz, and Todd Hester outline nine must-solve challenges before RL can safely power real-world systems. * Safety & constraints: avoid harmful actions while learning. * Sample efficiency: learn from limited,