OpenAI is expanding its Custom Model program to help enterprise customers develop custom generative AI models using its technology for specific use cases, domains, and applications.
Custom Model was launched last year at OpenAI's inaugural developer conference, DevDay, offering companies the opportunity to work with a group of dedicated OpenAI researchers to train and optimize models for specific domains. “Dozens” of customers have signed up for the custom form since then. But OpenAI says that by working with this initial group of users, it recognized the need to “evolve” the software to “maximize performance.”
It then helped with fine-tuning.
Assisted Fine-Tuning, a new component of the custom model software, leverages technologies beyond fine-tuning — such as “additional hyperparameters and various fine-tuning methods with broader parameter efficiency,” OpenAI says — to enable organizations to set up data-driven training pipelines and systems. Evaluation and more towards enhancing the model's performance on certain tasks.
OpenAI cites the example of SK Telecom, the Korean telecom giant, which worked with OpenAI to fine-tune GPT-4 to improve its performance in “telecom-related conversations” in Korean. Another client who helped with the fine-tuning, Harvey, which builds AI-powered legal tools with support from OpenAI's Startup Fund, collaborated with OpenAI to create a custom case law model that includes hundreds of millions of words of legal text and expert commentary. Lawyers.
“We believe that in the future, the vast majority of organizations will develop custom models tailored to their industry, business, or use case,” OpenAI wrote in a blog post. “With the variety of technologies available for custom model building, organizations of all sizes can develop custom models to achieve a clearer, more specific impact from their AI applications.”
Image credits: OpenAI
OpenAI is flying high, reportedly approaching a staggering $2 billion in annual revenue. But there are certainly internal pressures to maintain the pace, especially since the company is planning a $100 billion data center co-developed with Microsoft (if reports are to be believed). The cost of training and servicing leading generative AI models isn't going to come down anytime soon, after all — and consulting work like custom model training may be just the thing to keep revenue flowing reliably while OpenAI plans its next moves.
Along with the expanded Custom Model program, OpenAI today announced new model tuning features for developers working with GPT-3.5, including a new dashboard for comparing model quality and performance, and support for integrations with third-party platforms (starting with the AI Developer Platform) Weights and Biases ) and improvements to the tools. My mom talks about GPT-4 fine-tuning, which was released in Early Access during DevDay.
