AI Value Creators

Chapter 9. Generative Computing—A New Style of Computing

As you near the end of this book, you’re probably wondering: what’s next for LLMs? After all, large language models (LLMs) are undeniably peculiar creations, and even the experts (including us) can’t fully agree on what the future holds for this technology. The aim of this chapter, written with the help of a guest coauthor and VP of AI Models at IBM Research, David Cox, is to look into the future, with the nuances of the present, and introduce you to what we think will be a new style of computing that will take its rightful place with the other styles of computing we know today. In the previous chapter we discussed InstructLab, which anyone can use to contribute to training an LLM, akin to contributing to a software project. But what happens if we don’t just start building LLMs like they are software, but start building with LLMs like we build today’s software? Quite simply, today, people build with LLMs in an incoherent and unstructured messy way. We think those LLM-based applications need to be built in a structured, principled way, akin to how software is normally created. If this happens, there are some big benefits to be gained because software engineering principles like exception handling, buffer management, and more could all be applied to AI, which would help make models more efficient, safer, easier to work with, expressive, and more performant.

To us, it’s becoming apparent that LLMs aren’t going to be some set ...