AI Top-of-Mind for 6.20.24 — A More Private Cloud
Today: Apple’s Private Cloud Compute explained, a deep dive into CUDA, more China AI, and diversity from the Stanford AI Index Report.
Continuing week-after coverage of the Apple Intelligence launch, a detailed overview by Patrick Walsh writing in ‘The Salty Hash’ of the AI Private Cloud Compute (PCC) architecture. It is a combination of Apple hardware, a slimmed-down server OS, making sure data is ephemeral, anonymization, and even supply chain tamper resistance. The author points out some imperfections, but no first attempt is perfect!
If you are using Nvidia GPUs and CUDA and want to learn more about how it works under the covers, Daniel Warfield in ‘Towards Data Science’ offers a great (somewhat technical) tutorial:
· In this article we’ll use CUDA to train an AI model on a GPU, essentially implementing AI from scratch, assuming virtually no prior knowledge.
· First, we’ll explore some core components of modern computers, then we’ll dive into the GPU to describe what it is, how it works, and why it’s useful for AI. We’ll then work through an introduction to CUDA. We’ll describe what CUDA is and explain how it allows us to program applications which leverage both the CPU and GPU. Once we have an idea of how CUDA programming works, we’ll use CUDA to build, train, and test a neural network on a classification task.
Moving over to politics and China, ‘The Information’ looks at Chinese AI companies launching products and services in the US due to extreme price competition domestically. This in the face of increasing tensions and oversight. One company called out is Moonshot AI with its AI role-play and music video apps. From the article:
China’s crowded AI market is fueling the international ambitions of Chinese AI startups. Dozens of startups and large tech companies in the country have launched LLMs. Recently they have been undercutting each other’s prices in what could be a race to the bottom.
Revisiting the ‘Stanford AI Index Report,’ I’ll focus on diversity. The chart below pretty much says it all on STEM disparity, and the trends are even worse across Europe. I have a first-hand interest here with a daughter taking Electrical Engineering, and where she is in the minority in her class.
But a positive trend at the high-school level:
