More
    HomeAI Papers

    AI Papers

    The Entropy Enigma: Unlocking the Future of Reasoning in Language Models

    How Managing Policy Entropy Could Revolutionize Reinforcement Learning for LLMs Policy entropy collapse in reinforcement learning (RL) for large language models (LLMs) severely limits exploratory...

    One RL to See Them All: Unifying Visual Reasoning and Perception in AI

    V-Triune's innovative reinforcement learning system empowers vision-language models to master both complex thought and detailed sight, heralding a new era of versatile AI. Unified Training...

    Robin: Science with AI-Driven Discovery

    How a Multi-Agent System is Automating the Future of Therapeutic Innovation Robin, the first multi-agent AI system, fully automates the scientific discovery process by integrating...

    Virtual Worlds: CAST and the Future of 3D Scene Reconstruction

    Transforming Single RGB Images into Realistic 3D Environments with Component-Aligned Technology CAST (Component-Aligned 3D Scene Reconstruction) introduces a groundbreaking method to create high-quality 3D scenes...

    Fast Text-to-Audio Generation with Adversarial Post-Training

    Revolutionizing Audio Creation with Speed and Diversity Text-to-audio systems, despite their impressive performance, suffer from slow inference times, rendering them impractical for many creative applications,...

    AbsoluteZero: Transforming AI Reasoning with Self-Play and Zero Data

    How the Absolute Zero Paradigm Redefines Learning Without Human Input The Absolute Zero paradigm introduces a groundbreaking approach to AI reasoning, enabling large language models...

    INTELLECT-2 Breakthrough: Pioneering AI with Decentralized Intelligence

    The First 32B Parameter Model Trained Through Globally Distributed Reinforcement Learning INTELLECT-2 marks a groundbreaking achievement as the first 32-billion-parameter language model trained using globally...

    Continuous Thought Machines: Bridging the Gap Between Biological Brains and Modern AI

    How Neural Timing and Synchronization Could Unlock the Next Generation of Artificial Intelligence Biological brains rely on timing and synchronization for flexible, adaptive intelligence, but...

    SuperEdit: Image Editing with Smarter Supervision

    Transforming Instruction-Based Editing Through Rectified Guidance and Contrastive Learning SuperEdit introduces a groundbreaking approach to instruction-based image editing by rectifying editing instructions and aligning them...

    Apple’s Bold Step: How Length Skews Uncertainty Quantification

    Exploring the Hidden Flaws in UQ Evaluation and the Promise of LM-as-a-Judge Uncertainty Quantification (UQ) in Language Models (LMs) is vital for safety and reliability,...

    Values in the Wild: Uncovering the Hidden Judgments of AI

    How Language Models Like Claude Reveal Their Values in Real-World Conversations AI models like Claude, developed by Anthropic, are trained to reflect specific values such...

    WORLDMEM: Virtual Worlds with Lasting Memory

    How a Memory-Based Framework Ensures Long-Term Consistency in World Simulation WORLDMEM introduces a groundbreaking framework for world simulation, utilizing a memory bank to store past...