As businesses make more investments in AI, we’re going to start seeing more “custom AI” builds. “Custom AI” is the practice of taking a general foundation or large-language model (LLM) and adapting it so that it better reflects a particular organization’s needs. It all start with using that organization’s date, and then emplying tactics such as fine-tuning, continued pre-training, model-distillation, domain‐specific training, etc. to provide the most relevant and useful output. Think of a law firm using their own contracts to train their proprietary model, or a company using all of their own product specifications in the AI used for customer services.

Amazon AWS details this in a recent release: “Custom Intelligence: Building AI that matches your business DNA.” It outlines how they help customers create custom models and how these models can be fine-tuned over time.