What Is Learn-to-Steer? NVIDIA’s 2025 Spatial Fix for Text-to-Image Diffusion

As artificial intelligence transforms industries, universities must adapt or risk obsolescence. Institutions that embrace AI across various fields, enhance their computing infrastructure, and engage in innovative research will attract top talent and funding. This industry brief outlines a strategic framework for developing a comprehensive AI strategy that boosts enrollment and prepares graduates for future careers.

AI is revolutionizing various industries, but effective infrastructure is crucial for optimal performance. This ebook outlines strategies to enhance AI workloads, including optimizing infrastructure for applications like chatbots, utilizing dynamic batching and KV caching to reduce costs, and leveraging technologies like NVIDIA GPUs and Kubernetes for scalability.

The paper titled 'Quartet: Native FP4 Training Can Be Optimal for Large Language Models' discusses the advantages of training large language models (LLMs) directly in low-precision formats, specifically FP4. This method aims to reduce computational costs while enhancing throughput and energy efficiency. The authors introduce a new approach for accurate FP4 training, overcoming challenges related to accuracy degradation and mixed-precision fallbacks. Their findings reveal a new low-precision scaling law and propose an optimal technique named Quartet.

Microsoft and NVIDIA have announced plans to invest up to $15 billion in Anthropic, an AI safety and research company. This investment aims to enhance the development of advanced AI technologies while ensuring safety and alignment with human values. The collaboration is expected to leverage Anthropic's expertise in AI safety to create more robust and responsible AI systems.

Microsoft and NVIDIA have announced a joint investment of up to $15 billion in Anthropic, an AI safety and research company. This investment is part of a broader strategy to enhance the development of advanced AI technologies. Anthropic has also committed to purchasing $30 billion in Azure compute capacity, with the option for additional resources, indicating a significant partnership aimed at advancing AI capabilities while ensuring safety in deployment.

MeanFlow (MF) is a generative model inspired by diffusion processes, designed for efficient few-step generation by learning direct transitions from noise to data. It is commonly utilized as a latent MF, employing the pre-trained Stable Diffusion variational autoencoder (SD-VAE) for high-dimensional data modeling. However, MF training is computationally intensive and often unstable. This study introduces an efficient training and sampling scheme for MF in the latent space of a Representation Autoencoder (RAE), addressing issues like gradient explosion during training.

The article discusses the development of a portable and efficient large language model (LLM) inference platform using a state-of-the-art paged attention kernel. This kernel, built on the Triton language, aims to enhance performance on both NVIDIA and AMD GPUs without requiring low-level hand-tuning. The authors detail their approach, algorithmic improvements, and the necessary auto-tuning for efficiency, achieving a significant performance increase from 19.7% to a much higher benchmark.

Data attribution for text-to-image models seeks to identify the training images that significantly influenced generated outputs. Current methods require substantial computational resources for each query, limiting their practicality. A novel approach is proposed for scalable and efficient data attribution, distilling a slow, unlearning-based method into a feature embedding space for quick retrieval of influential training images. The method, combined with efficient indexing and search techniques, demonstrates competitive performance on medium and large-scale models, achieving results faster th…