Serverless Composability

ChatGPT took the world by storm with the quality of its answers. However, the true power of LLMs is their ability to use tools and work with each other.

Tools

A model can become more capable by learning how to use tools. Because they are experts at natural language, so they can understand documentation in the same way a human can e.g. GPT-4 Plugins.

Amazingly, models can even learn how to use other models.

Tools are beneficial for the owner of the model because they can build an app that uses the more capable model. If the model is closed source, it needs to have a public API for third-party prompt creators to use it and their app becomes dependent on that API.

Composability

Composing models is a more flexible approach because the product is not dependent on any one model. If a model API is turned off or a better model is released, then that capability can be replaced with a model that provides a similar function.

Many of the most valuable use cases of LLMs combine multiple models. By default, the workflow involves manually copying the output of one model to another. This works to create a single artifact, but not a product that can be used repeatedly.

Langchain lets engineers chain together models, giving them more control over the user experience. However, going from a chain to an end-to-end GenAI product such as MakeLogo is still a significant engineering project.

PromptSpace

PromptSpace introduces Serverless Composability. Combine models by describing your app in PSL.