Creating Custom GPTs
The Myths and Realities
I have seen authors hype the capacity of custom GPT models. Custom GPTs are attractive because they are tailored to suit individual or organizational needs. For instance, a writer may input a variety of articles they have written. Then, ask the AI to generate a detailed analysis of the writing and output a sequence of statements that define their personal writing characteristics. These statements are used to create a custom GPT to write prose similar to the writer. Essentially, the writer is cloning themselves.
Custom GPTs have captured the imaginations of many AI applications reaching into industries from business to research; writers and enthusiasts alike have been touting the idea that anyone can craft a bespoke version of GPT (Generative Pre-trained Transformer) with seemingly limitless capabilities. Unfortunately, much of this excitement is based on misconceptions about what fine-tuning large language models (LLMs) like GPT-4 can actually achieve.
The Reality of Fine-Tuning Large Language Models
It’s important to understand that a custom GPT model is not a new model built from the ground up. When you fine-tune an existing model like GPT-4, you’re starting with a pre-trained model that already has knowledge of vast amounts of text — terabytes of data…
