NVIDIA has unveiled its most affordable generative AI supercomputer, “Jetson Orin Nano Super Developer Kit”.

OpenAI introduces demo developers and updates APIs.

Arm Holdings CEO Rene Haas discusses the rapid evolution of AI and its impact on the chip industry, emphasizing Arm’s strategic role in providing architectures for a wide range of applications. He addresses Arm’s potential move into designing AI chips, Intel’s challenges, and the importance of balancing innovation with geopolitical considerations, especially in China. Haas also highlights Arm’s growing presence in AI workloads across various devices, stressing the need for more compute capabilities to support AI’s expansion.

Giraffe introduces a transformer-based architecture capable of processing significantly longer input contexts, pushing the boundaries of context length in open-weight models.

Meta researchers have developed Adaptive Caching (AdaCache), a training-free method designed to accelerate the video generation process for Diffusion Transformers.

The Digital Typhoon Dataset is the longest satellite image dataset for typhoons, spanning over 40 years.

Wondering how product managers should use AI in their work? Take Pendo and Mind the Product’s free online course, AI for Product Management, to learn core AI and product use cases—and earn a badge that you can add to your LinkedIn profile.

Register for free ->

RetroLLM is a framework designed for training large language models with retrieval augmentation, enabling more efficient and accurate responses by integrating external knowledge sources during inference.

Iteration of Thought (IoT) uses dynamic, thought-provoking prompts to improve LLMs. Unlike existing methods like Chain of Thought (CoT), IoT adapts to the context of each interaction.

TokenFormer is a novel architecture designed to reduce the high computational costs associated with scaling transformer models.

Grok has a new version and a new efficient model that is available for all users. It also has an improved image generation model and API.

World models enable AI systems to simulate and reason about their environments, pushing forward autonomous decision-making and real-world problem-solving.

AI is useful for tasks like idea generation, summarization, and translation where quantity or varied perspectives are needed. It excels when humans can quickly assess AI’s output and in tasks where minimal downside exists for errors. AI use can hinder learning or accuracy in high-stakes or unfamiliar situations, requiring careful judgment to balance its utility.

Grok’s new image generation model excels in rendering real-world details, logos, and realistic human portraits.

YouTube’s auto-dubbing feature is now available to hundreds of thousands more channels, focusing initially on informational content.

Google signed a $20 billion deal with Intersect Power and TPG Rise Climate to build renewable energy and storage for its data centers.