/
Customized large language models could make generative AI more useful for the business customers OpenAI is pursuing.
Share this story
:format(webp)/cdn.vox-cdn.com/uploads/chorus_asset/file/24390407/STK149_AI_02.jpg)
Illustration: The Verge
OpenAI’s enterprise customers who want to use generative AI but prize accuracy might be able to get that with new customization options for the GPT-4 API.
They include the ability to connect to third-party platforms to share information on fine-tuning, save fine-tuned models during additional training without retraining the entire model, and a new user interface to compare model performance and quality. The OpenAI API was launched in 2020 to allow companies to connect and build on top of its large language models, which was followed last August by opening fine-tuning for GPT-3.5 and eventually GPT-4, too.
Highly customized models could provide better results internally. They could also keep responses to the public on track and avoid embarrassing situations like the ChatGPT-powered Chevy dealership’s bot that said it would sell a car for $1.
Today, OpenAI COO Brad Lightcap said in a Bloomberg interview that 2024 is the “year of the enterprise” as companies begin to come out with more use cases using foundation models and that OpenAI is seeing “tremendous growth” in the space. “We feel increasingly that the market is pulling us toward real applications, delivering real business results with a really broad focus on AI enablement, and we’re ready to support our customers with that,” Lightcap said.
Lightcap previously warned executives not to expect AI to solve all of their problems and cautioned that generative AI is still very much in the experimental stages.
OpenAI also officially announced assisted fine-tuning, where OpenAI employees help customers fine-tune GPT-4 as a part of its Custom Models program. The Custom Models program allows selected organizations to work with OpenAI researchers to train custom GPT-4 models to their specific needs.
Fine-tuning allows pretrained models like GPT-4 to learn more information through smaller, more targeted datasets like a company’s portfolio. The first third-party integration is with the platform Weights & Biases.
While OpenAI is mostly known for ChatGPT, which attracts both everyday consumers and brands that want to create their own versions of the chatbot, the company has worked with startups and more established organizations to bring AI-powered applications and services to the mainstream.
Some of the earliest darlings of the generative AI age, such as copywriting platform Jasper and video creation company Descript, built their services using OpenAI’s model.