ChatGPT, DALL-E, and More: How Generative AI Works and How It Is Used For Science

The releases in the past two years of ChatGPT, DALL-E and similar tools have a profound impact on the public’s perception of and access to powerful AI capabilities. While the entire fields of Machine Learning (ML), Artificial Intelligence (AI) and Natural Language Processing (NLP) have advanced tremendously in recent years, no other developments therein have prompted quite the level of attention, investment, disruption, and backlash that these text- and image-generation models have managed in a very short time. ChatGPT belongs to a class of ML methods called Large Language Models (LLMs) which are seeing rapidly-increased usage in industry and research. DALL-E and similar image generators are built in a variety of ways, including with diffusion models. Already, numerous projects across BNL and elsewhere in the scientific community are exploiting these tools, and their scientific reach and scope appears likely to continue growing. In this talk, I will provide background on the fields of AI and ML, describe the techniques that make generative AI possible, provide further details about ChatGPT and DALL-E specifically, and discuss the applications and impact these methods are having in science and the broader world.

Topical Areas: Computer Science; Data Science; Artificial Intelligence

Topic/Methods/Domain: This seminar describes the AI and Machine Learning methods that makes powerful tools like ChatGPT and DALL-E possible, including deep convolutional neural networks (CNNs), Transformers, Generative Adversarial Networks (GANs) and others.

Target Audience: Undergraduate students; master’s students; Ph.D. students

Bio:
Dr. Carlos X. Soto was born in San Luis de La Paz, a small town in central Mexico, and grew up as the second of six siblings in San Antonio, TX. He is a first-generation student (college or otherwise) and attended Texas A&M University where he
received his Bachelor of Science and Ph.D. degrees, both in Computer Engineering. His graduate research focused on perception algorithms, collaborative interfaces, and human-robot interaction for search and rescue robotics. In 2018 Dr. Soto joined what would become the AI and Machine Learning department at Brookhaven National Laboratory, where he now leads the AI Theory and Security research group. He works on problems in machine learning and natural language processing, with applications in biology, nuclear physics, radiochemistry, and nuclear nonproliferation. Outside of work, he is passionate about nature, film, technology, and knowledge for its own sake.

https://www.bnl.gov/staff/csoto