OpenAI expands its custom model training program

OpenAI is expanding a program called Custom Models to help enterprise customers use its technology to develop customized generative AI models for specific use cases, domains and applications.

Last year, custom models were launched at DevDay, OpenAI’s first developer conference, providing companies with the opportunity to work with a dedicated group of OpenAI researchers to train and optimize domain-specific models. Since then, “dozens” of customers have signed up for custom models. But OpenAI said that while working with this initial group of users, it came to realize that it needed to “evolve” the program to “further maximize performance.”

therefore Auxiliary fine-tuning.

Assisted fine-tuning is a new component of the custom model program that leverages technology Exceed Fine-tuning—such as, in OpenAI’s words, “larger scale additional hyperparameters and efficient fine-tuning methods for various parameters”—enables organizations to build data training pipelines, evaluation systems, and more to enhance model performance on specific tasks.

OpenAI cited the example of South Korean telecom giant SK Telecom, which worked with OpenAI to fine-tune GPT-4 to improve its performance in Korean “telecom-related conversations.” Another assisted fine-tuning client, Harvey, which is building an AI-driven legal tool with support from the OpenAI Startup Fund, worked with OpenAI to create a custom model of case law containing hundreds of millions of words of legal text and feedback from expert attorneys.

“We believe that in the future, the vast majority of organizations will develop models that are individually tailored to their industry, business, or use case,” OpenAI wrote in a blog post. “With multiple technologies available for building custom models, organizations of all sizes will Organizations can develop personalized models to achieve more meaningful and specific impact from their AI implementations.”

Image Source: open artificial intelligence

OpenAI is growing rapidly, with annual revenue reportedly approaching $2 billion. But there’s certainly internal pressure to keep pace, especially as the company plans to develop a $100 billion data center with Microsoft (if reports are to be believed). After all, the cost of training and servicing flagship generative AI models isn’t going down anytime soon, and consulting work like custom model training might just be the way to keep revenue flowing reliably as OpenAI plots its next steps.

In addition to its expanded custom model program, OpenAI today announced new model fine-tuning capabilities for developers using GPT-3.5, including a new dashboard for comparing model quality and performance, support for integration with third-party platforms (from AI Developer platform begins) weights and biases) and tool enhancements. Mom fine-tuned GPT-4, which was launched into early access during DevDay.