ChatGPT wants to know more about your company.
OpenAI will now allow businesses to create their own custom versions of GPT-3.5 Turbo, the A.I. model that powers ChatGPT. If you've been fantasizing about combining generative A.I. with your company's proprietary data, GPT-3.5 Turbo could potentially be a game changer.
The process of customizing a pre-existing A.I. model, also known as fine-tuning, consists of training the model on a new set of data specifically chosen to enable narrower use cases.
Imagine that you are the owner of a small brewery, and you want to get the word out about your brand with an A.I.-powered chatbot to engage potential consumers on social media. You could train GPT-3.5 Turbo on data related to your craft beers, enabling the chatbot to answer specific questions and make recommendations. You could also use fine-tuning to develop a unique "brand voice" for the chatbot with a more defined personality than the base model. OpenAI says all fine-tuning data is private, and won't be used to train the company's publicly available models.
Fine-tuning allows businesses to shorten their prompt lengths without sacrificing performance. OpenAI's suite of APIs charge users per character in both the prompt and the chatbot's response. With fine-tuning, companies are also charged depending on the length of the training data. That might seem like a bad deal, but the company says testers of the fine-tuning feature "reduced prompt size by up to 90 percent by fine-tuning instructions into the model itself, speeding up each API call and cutting costs."
So how can you get started? The first and most important step is collecting and formatting your training data. Every piece of data that you want to feed to the A.I. needs to be formatted as an interaction between a user and the model, so try to imagine how a typical user of your custom A.I. will talk with the model, and write up as many example conversations as you can possibly think of. This means you'll need to convert all of your data into a conversational format.
If you need some help getting your data correctly formatted, OpenAI has named Scale, a firm that works with businesses to build custom A.I. models, as a "preferred partner" for fine-tuning GPT-3.5 Turbo. By using Scale's "Data Engine" tool, OpenAI says you can automatically convert your data into the correct format for fine-tuning.