OpenAI is expanding its program Custom Models to enable enterprise customers to develop customized generative AI models using the company's technology for specific use cases, domains, and applications.
Custom Model was announced last year at DevDay, OpenAI's first developer conference, offering companies the opportunity to work with a group of dedicated OpenAI researchers to train and optimize models for specific domains. Since then, “dozens” of customers have signed up for the custom model. However, OpenAI says that while working with this initial group of users, it began to recognize the need to scale the program to further “maximize performance.”
Therefore, fine-tuning and custom-trained models came to the rescue.
Assisted fine-tuning, a new component of custom model programs, takes techniques beyond fine-tuning, including, in OpenAI's words, “additional hyperparameters and efficient ways to fine-tune a large variety of parameters.” Data training pipelines, evaluation systems, and other supporting infrastructure that organizations can leverage to enhance model performance on specific tasks.
For custom training models, OpenAI uses OpenAI's base models and tools (such as GPT-4) for customers who need to fine-tune their models more deeply or incorporate new domains. A custom model built. “We have specific knowledge,'' he says of OpenAI.
OpenAI cites the example of South Korean telecom giant SK Telecom. SK Telecom worked with OpenAI to fine-tune his GPT-4 to improve its performance in “telecommunications-related conversations” in Korean. Another customer, Harvey, whose AI-focused venture arm at OpenAI is building AI-powered legal tools with support from the OpenAI Startup Fund, is working with OpenAI to generate billions in billions of We created a custom model of case law that incorporates legal documents from the same language. Feedback from qualified professional lawyers.
“In the future, we believe the majority of organizations will develop models customized to their industry, business, and use case,” OpenAI wrote in a blog post. “The variety of techniques available for building custom models allows organizations of all sizes to develop personalized models to achieve more meaningful and tangible impact from their AI implementations.”
Image credit: OpenAI
OpenAI is growing by leaps and bounds and is reportedly approaching an impressive $2 billion in annual revenue. But there's certainly internal pressure to keep up the pace, especially since the company is planning a $100 billion data center jointly developed with Microsoft (if reports are to be believed). After all, the cost of training and delivering flagship generative AI models isn't going to drop anytime soon. While OpenAI plans its next move, consulting work, such as training custom models, may be needed to continue growing revenue.
Fine-tuned custom models can also reduce strain on OpenAI's model serving infrastructure. Customized models are often smaller and more performant than generic models, making them an attractive solution for OpenAI, which has historically had computational power challenges, as demand for generative AI reaches its peak. There is no mistake.
In addition to the expanded custom model program and custom model building, OpenAI today announced new dashboards for developers using GPT-3.5 to compare model quality and performance, and integration with third-party platforms. Support for the AI developer platform Weights & Biases and tool enhancements. However, mom's word for his GPT-4 tweaks launched in early access during DevDay.