Skip to content
Tarun Bevara

My Articles

  • Why Do LLMs Get Smarter as They Get Bigger? The Hidden Geometry of “Superposition”
    2025

    Why Do LLMs Get Smarter as They Get Bigger? The Hidden Geometry of “Superposition”

    Read Article
  • Beyond Diffusion: How Visual Autoregressive Modeling (VAR) Rewrites the Rules of Image Generation
    2025

    Beyond Diffusion: How Visual Autoregressive Modeling (VAR) Rewrites the Rules of Image Generation

    Read Article
  • Breaking the Transformer Bottleneck: Gating for Scaled Non-Linearity and “Attention-Sink-Free” LLMs
    2025

    Breaking the Transformer Bottleneck: Gating for Scaled Non-Linearity and “Attention-Sink-Free” LLMs

    Read Article
  • Beyond RAG: Why CLaRa’s “Latent Reasoning” Changes the Game for Retrieval-Augmented Generation
    2025

    Beyond RAG: Why CLaRa’s “Latent Reasoning” Changes the Game for Retrieval-Augmented Generation

    Read Article
  • KAN: The Shift from Fixed Nodes to Learnable Edges
    2025

    KAN: The Shift from Fixed Nodes to Learnable Edges

    Read Article
  • The Magic Number is 3.6: Measuring the True Capacity of Large Language Models
    2025

    The Magic Number is 3.6: Measuring the True Capacity of Large Language Models

    Read Article
  • Beyond the Click: Inside the Architecture of SAM 3 and Promptable Concept Segmentation
    2025

    Beyond the Click: Inside the Architecture of SAM 3 and Promptable Concept Segmentation

    Read Article
  • We’ve Been Overcomplicating Diffusion: Why “Just Image Transformers” Are All You Need
    2025

    We’ve Been Overcomplicating Diffusion: Why “Just Image Transformers” Are All You Need

    Read Article
  • Nested Learning: A Technical Reflection on a New Way to Think About Deep Learning
    2025

    Nested Learning: A Technical Reflection on a New Way to Think About Deep Learning

    Read Article
  • The Transformer Paradigm: Rethinking Sequence Modeling Through Attention
    2025

    The Transformer Paradigm: Rethinking Sequence Modeling Through Attention

    Read Article
  • Inside PaliGemma: Building an Open, Transferable 3B Vision-Language Model
    2025

    Inside PaliGemma: Building an Open, Transferable 3B Vision-Language Model

    Read Article
  • Less Is More: My Deep Dive into Recursive Reasoning with Tiny Networks
    2025

    Less Is More: My Deep Dive into Recursive Reasoning with Tiny Networks

    Read Article