OpenAI Expands Custom Model Training Program

OpenAI is expanding its Custom Model program to assist enterprise customers in developing tailored generative AI models for specific use cases, domains, and applications. This initiative, which was launched last year at OpenAI’s developer conference, DevDay, has seen enrollment from “dozens” of customers. Building upon this initial success, OpenAI is introducing new components to the program to further optimize model performance.

One of these components is assisted fine-tuning, which goes beyond traditional fine-tuning techniques by incorporating additional hyperparameters and parameter-efficient methods at a larger scale. This approach allows organizations to establish data training pipelines and evaluation systems to enhance model performance on specific tasks.

Another aspect of the expanded Custom Model program involves creating custom-trained models using OpenAI’s base models and tools, such as GPT-4. These custom models are tailored for customers who require deeper fine-tuning or domain-specific knowledge integration. For example, SK Telecom collaborated with OpenAI to fine-tune GPT-4 for telecom-related conversations in Korean, while Harvey, a company developing AI-powered legal tools, worked with OpenAI to create a custom model for case law using vast amounts of legal text and expert feedback.

OpenAI believes that customized models personalized to industries, businesses, or specific use cases will become increasingly prevalent in the future. By offering a range of techniques to build custom models, organizations of all sizes can develop personalized AI solutions to achieve more impactful results in their implementations.

In summary, the expanded Custom Model program aims to empower enterprises to harness the full potential of generative AI technology by tailoring models to their unique requirements and objectives.