OpenAI is stepping up its efforts to help businesses develop customized AI models by expanding its Custom Model program and introducing assisted fine-tuning capabilities. This comes as the company's enterprise-focused product, ChatGPT, has seen rapid adoption, with the number of registered ChatGPT Enterprise users increasing fourfold since January to over 600,000, according to Chief Operating Officer Brad Lightcap. The Custom Model program was initially launched at OpenAI's first DevDay developer conference last year and has since attracted dozens of clients who want to collaborate directly with OpenAI researchers to train and optimize models for their unique needs. The success of the program has prompted OpenAI to expand its services to better meet the demands of its enterprise customers. In August 2023, OpenAI introduced a self-serve fine-tuning API for GPT-3.5, which has been used by thousands of organizations to train hundreds of thousands of models. The new features of the fine-tuning API include checkpoint creation based on epochs, a new side-by-side Playground UI for comparing model quality and performance, integration support with third-party platforms (starting with Weights and Biases this week), comprehensive evaluation metrics, and configurable hyperparameters from the dashboard. A key addition to the program is assisted fine-tuning, an advanced technique that goes beyond the standard fine-tuning API and involves collaboration between the OpenAI technical team and clients. This includes the use of additional hyperparameters and efficient parameter-efficient fine-tuning (PEFT) methods at a larger scale. Assisted fine-tuning is particularly beneficial for organizations that require support in building efficient training data pipelines, evaluation systems, and custom parameters to maximize the performance of their models for specific use cases or tasks. SK Telecom, a telecommunications operator in South Korea, has seen significant improvements in its customer service AI through assisted fine-tuning. By collaborating with OpenAI and fine-tuning GPT-4 for telecommunications-related conversations in Korean, SK Telecom achieved a 35% improvement in conversation summarization quality, a 33% increase in intent recognition accuracy, and a significant increase in satisfaction ratings compared to the base GPT-4 model. For organizations with highly specific use cases and large amounts of proprietary data, OpenAI offers the option to train fully customized models. These models are built from scratch and incorporate domain-specific knowledge through novel techniques in mid-training and late-training. Harvey, an AI-native legal tool for lawyers, has partnered with OpenAI to create a large-scale language model tailored to case law. By modifying every step of the model training process and incorporating the equivalent of 10 billion tokens of legal data, the final model achieved an 83% improvement in factual responses, with lawyers preferring its outputs in 97% of cases over GPT-4. OpenAI believes that the future of AI lies in the development of customized models for specific industries, businesses, and use cases. From the self-serve fine-tuning API to the Custom Model program, organizations of all scales can now create personalized AI models to drive more meaningful and specific impacts in AI implementation. As the demand for customized AI solutions continues to grow, OpenAI's expansion of the Custom Model program and introduction of assisted fine-tuning will play a crucial role in helping businesses fully leverage the potential of AI technology.