Mar 30, 2025

• Claude 3.5 Haiku demonstrates advanced capabilities, such as multi-step reasoning and poetic planning, revealing its internal processes through attribution graphs, which enhance our understanding of model behavior.

• Certified randomness was achieved using a 56-qubit trapped-ion quantum processor, demonstrating the potential for practical applications in cryptography and secure communications through a novel protocol that combines quantum and classical computing methods.

• The first LLM, according to Jeremy Howard, is ULMFit, a self-supervised LSTM model trained on Wikipedia, which laid the groundwork for subsequent models like GPT-1 that utilized the transformer architecture for enhanced performance.

• DeltaProduct enhances state-tracking in linear RNNs by utilizing multiple steps of online gradient descent per token, leading to improved expressivity without sacrificing efficiency.

• Lumina-Image 2.0 introduces a unified transformer-based architecture that efficiently handles text-to-image generation, image editing, inpainting, and outpainting, utilizing a novel Multiple Sampling with Iterative Refinement (MSIR) technique to enhance image quality without added computational costs.

• TextGrad introduces a novel framework that optimizes generative AI by backpropagating feedback from large language models (LLMs), enabling automatic improvements across various tasks, including science problem-solving and treatment planning.

• UI-R1 innovatively combines rule-based reinforcement learning with large language models to enhance GUI agents, enabling them to learn from mistakes and adapt to new interfaces effectively.