TLDR AI 2024-08-08

Google announced new Gemini AI-powered features for Google Home, including intelligent captions for Nest camera footage, natural language processing for Home routine creation, and an upgraded, more natural-sounding Google Assistant. The advanced features, primarily behind a Nest Aware subscription paywall, aim to enhance the smart home experience with the rollout commencing in beta and expanding next year. As part of the move toward smarter home automation, Google envisions an assistant that can proactively manage complex and dynamic home environments.

Nvidia is facing increased government scrutiny from the EU, UK, China, and the US Justice Department over its dominant market share in AI chips and sales practices. The company is rapidly building its legal and policy teams to address antitrust concerns amid profitable growth, as it commands 90 percent of the GPU market essential for AI systems. Nvidia is also adapting to increased competition oversight, with recent attention turning to its planned acquisition of Run.ai and impact on the AI supply chain.

Meta plans to significantly increase computing capacity for training its next-gen large language model, Llama 4, expecting a 10x compute increase over Llama 3. The investment in AI training infrastructure will drive up capital expenditures in 2025. Despite heavy spending, Meta doesn't foresee immediate significant revenue from Gen AI products.

There is strong pressure to use compute at inference time to improve model performance. This paper showcases several methods that can be used and discusses the trade offs made between them. In general, this points towards a broader trend of squeezing performance out of smaller models.

Treating 3D objects as UV wrapped images makes it simple to use a DiT model to synthesize novel objects based on textual inputs.

There has been a push in vision language models to make them easy to tune and transfer between tasks. This is reminiscent of computer vision a decade ago. Importantly, this can be done through careful data curation and architecture updates which are on display with LLaVA-OneVision.

XLabs has released the first Flux-Dev control net which allows for generation conditioned on Canny image inputs.

HARMONIC is a framework for generating and evaluating synthetic tabular data using large language models.

muP is a way to tune your hyperparameters to transfer from smaller to larger models. This GitHub gist shows a great theorem in practice which states that you can change where you scale model outputs and it doesn't affect final transfer performance.

GitHub is leveraging AI and machine learning to analyze and summarize customer feedback at scale, informing product improvements and feature prioritization with actionable insights. This automated approach enables the capture of multilingual feedback and drives data-driven decision-making, enhancing the responsiveness to developer needs. The initiative highlights GitHub's commitment to integrating AI to maintain a developer-centric product development process.

This article presents a deep skepticism about OpenAI's sustainability, considering generative AI's lack of broad market utility and the prohibitive costs of developing and running large language models. The author questions OpenAI's long-term viability without significant technological breakthroughs or continuous, unprecedented fundraising efforts. Despite OpenAI's influence on the AI industry, the company faces challenges with profitability, high operational burn rates, and dependence on strategic partnerships, particularly with Microsoft.

AMD's Q2 2024 earnings show a significant shift toward data center products, with nearly half of its sales now in this area, thanks primarily to the Instinct MI300 AI chip. The company has committed to annual new AI chip releases, competing with Nvidia's offerings, despite supply constraints projected to last until 2025. Although Nvidia remains ahead in the data center market, AMD has seen growth in its CPU and GPU segments, including its Ryzen processors and Radeon 6000 GPUs.

Yum! Brands is expanding its AI-driven Voice AI technology to hundreds of Taco Bell drive-thrus to enhance efficiency, aiming for broader rollout by year-end.

For human-like AI voice generation, models need to be trained on diverse pronunciation data. Discover more with this on-demand webinar hosted by Oxford Languages and WellSaid. Watch the video

OpenAI is leading a $60 million funding round for Opal, the same company behind the high-end Tadpole webcam.

Microsoft's AI investments surged to $19 billion last quarter, focusing on data centers for AI capabilities, but significant AI revenue remains uncertain.