Jon Durbin
Self-instructing AI innovation pioneer
About
Jon Durbin is a distinguished figure in the AI community, particularly known for his contributions to generative AI and Large Language Models (LLMs). Based in the Detroit Metropolitan Area, Durbin operates as a self-employed software engineer and AI researcher. His extensive experience as a senior applied AI researcher at Convai highlights his expertise in developing AI-powered non-player characters for video games. His academic foundation was laid at the University of Michigan-Flint, leading to a prolific career characterized by a deep-seated interest in various facets of AI. Durbin has made significant strides in AI research focused on fine-tuning models and curating datasets. He is renowned for the creation of the Airoboros dataset and associated models, pioneering a paradigm that utilizes AI-generated data for training, which often rivals the quality of human-generated data models. These models are accessible on Hugging Face, a testament to his commitment to the open-source community. Durbin's innovations don't stop there; his work on the Bagel models has set new standards in model development, emphasizing the enhancement of factual accuracy and mathematical prowess through Direct Preference Optimization (DPO). The Bagel models, inspired by the film "Everything Everywhere All at Once," showcase his unique approach to integrating diverse datasets and techniques. This reflects a broader exploration within the AI landscape of merging models to develop larger, more capable systems without the need for extensive retraining. Durbin's work is continually referenced and debated within the AI community, owing to its transformative impact. Furthermore, he is actively engaged with peers, sharing insights on platforms like LinkedIn and Buy Me a Coffee, thereby fostering a collaborative environment for AI advancements. Durbin is not only a researcher but also a thought leader, aiming to expand his digital presence through blogging and video content, further amplifying his influence in the AI domain.
