Foundational Generative AI Models Are Like Operating Systems

By Bill Franks June 18, 2024

Collected at: https://datafloq.com/read/foundational-generative-ai-models-are-like-operating-systems/

With generative AI evolving so rapidly, there aren’t many things that can be stated with confidence about exactly where it will go in the future. This blog will address one concept that I am convinced will hold true regardless of how generative AI evolves. Namely, foundational generative AI models will play a role very similar to classic operating systems and will effectively serve as commoditized utilities.

Operating Systems Galore? Not So Much…

Every computer and mobile phone runs on an operating system. If you have a laptop, it is almost certain that you are running either Microsoft Windows or Apple MacOS. If you have a phone, it is almost certain you’re running either Google Android or Apple iOS. While other operating systems do exist, the vast majority of users have settled on one of a limited set of options. Limited options dominate because they each incorporate a massive range of features that cover all common requirements.

While there are some differences in how Windows and MacOS operate, most users can do the same things on either one. The same applications are available for each operating system and the user interfaces are similar. Phone operating systems are even more alike. It is hard to find a major difference in how to operate an Android phone versus an iOS phone. On top of that, virtually every app in existence can run on either platform. These operating systems are all effectively foundational, commodity platforms that you pay a fee to make use of … like a utility service.

Nobody Creates A Custom Operating System

We never hear of companies investing in their own operating systems because the robust functionality of standard operating systems meets their needs. The amount of money, time, and opportunity cost required to create a legitimate alternative to today’s dominant operating systems is simply too high to make it worthwhile for anyone to build their own.

What organizations (and individuals) do instead is to differentiate based on the applications they lay on top of their operating system. Computers and phones with different owners have vastly different assortments of applications on them, and the owners of the devices might use them for drastically different purposes. So, while the operating systems are largely commoditized and interchangeable, the variety of ways they can be put to work for any given user is almost unlimited.

Why Foundational Generative AI Models Are Like Operating Systems

There are a handful of widely used foundational generative AI models available today. OpenAI, Google, and Meta have three of the most common. Like operating systems, there are some differences in the strengths and weaknesses of these offerings. However, they are fundamentally built on the same type of model architecture, using the same type of training data, and are targeted for the same uses. Moreover, they require a massive and expensive training data collection process and a mind-blowing level of cost for model development. Maintaining the models and letting users query them isn’t cheap either!

Thus, we can expect to see just a few foundational generative AI models survive and become the standard bearers. Like with computer and phone operating systems, it just won’t make sense to build your own. Rather, we’ll pay a fee to a foundational model owner to tie into its capabilities much as we pay a fee to Microsoft, Apple, or Google for the ability to use their operating systems … and much like we pay a fee to access utility networks offering power or natural gas service.

How To Differentiate While Using The Same Foundational Models

The standardization around a few foundational models doesn’t mean that companies and individuals are limited in how they can differentiate themselves and create value with generative AI. Do software companies feel limited in their ability to differentiate when writing a Windows application? No! The differentiation is in how the underlying operating system is used to support each application. There are unlimited opportunities for creators to drive innovation and value with their applications even though they are built upon standard operating systems.

The same will happen with generative AI. While there may only be a few foundational models to choose from, they will be robust and easy to build on top of. We’ll continue to see an explosion of innovation and new revenue streams derived from generative AI. The foundational models will be commoditized utilities that support those efforts rather than being the primary focus.

So, go out and drive value with generative AI. Just make sure that you’re focused on building on top of a foundational model instead of trying to build your own. Pay the model owner a fee for your access, of course. Then, use that foundational model like an operating system by focusing on adding value through the applications and processes that you build on top of it!