The Business Architecture of AI

FourWeekMBA

The way the AI industry is evolving right now:

• Closed-Centralized-top-down data processing: with the OpenAI-Microsoft partnership, Microsoft Azure offers the cloud infrastructure which enables OpenAI to pre-train its powerful models (right now, GPT-4 might be getting trained on trillions of parameters that require a huge amount of computation). And an Open APIs infrastructure where any business can access the OpenAI APIs to build business applications. Here you can’t download the model, and each iteration of your build will go through the foundational layer built by OpenAI APIs. So as you train and fine-tune a middle layer model on top of GPT-3, you’re interacting with OpenAI.
• Open-Decentralized-On edge data processing: with Stable Diffusion getting pre-trained on AWS, once the model is ready, it can get downloaded by anyone and run on a device, like the iPhone and on edge. Meaning you download and use it on your device without the data needing to go back to the centralized cloud provider. The reason why here I say it's open, it's because you can download the full, open-sourced version once the model has been released. And you don't need to interact with its APIs, but you can download Stable Diffusion and run it, let's say, on your iPhone. And therefore, it's on edge because there is no data that goes back to Stability AI, which develops Stable Diffusion. You can use it on your device at a local level.