Large language models are modern day operating systems

The market for LLMs will evolve in the same way that the market for traditional operating systems did

AI models strut their stuff on a (artificial?) catwalk.

One way to think of large language models (LLMs) is as operating systems for computational intelligence. Some quick, brainstorming thoughts follow, about what could arise, given this framing.

Platform wars: Just like the operating system market, we’ll see platform wars emerge between the dominant LLM providers (OpenAI, Google, etc.) This will lead to a battle for developer mindshare, ecosystem lock-in, and proprietary standards & APIs. In many ways this will be a recpitulation of the Windows/DOS wars of the ‘80s and early ‘90s.

Compatibility & Interoperability: Compatibility and interoperability will become crucial. Think of sharing files between Windows and Mac users. This used to be an issue, then as the market forced the two companies to complement each other, it became a solved problem. What about ‘virtual machine’ layers or translation tools that allow apps built on one LLM to run on others?

Open-source movement: We’re seeing this already with Mistral, etc., similar to the rise of Linux. Academic institutions, research labs (ex. OpenAI, to the extent it’s still considered a research lab) will probably lead this development. These models may cater to niche use cases or serve as alternatives to proprietary offerings.

Specialized LLMs: Similar to specialized operating systems for embedded devices. For example, a specialized language model for legal work, healthcare, finance, or scientific research.

Ecosystem & Developer Tools: As the LLM market matures, we will see developer tools, frameworks, and libraries all built to allow LLM-powered apps and services.

Monetization & Business Models: LLM providers will monetize in different ways, including subscription-based models, pay-per-use, or value-added services and tools.

Integration & Hybrid Approaches: Integration of LLMs with other technologies, such as traditional software systems, databases, and domain-specific models.