|
STATISTICS AND LARGE LANGUAGE MODELS
WORKSHOP SCHEDULE MONDAY, JULY 24, 2023 |
|
1:00 - 1:45 PM
Language Models: From N-Grams to Transformers to Chatbots Presentation Slides https://www.nycasa.org/ASA-LLM-Workshop-Jul-24-2023.pdf by Bob Carpenter Flatiron Institute Abstract Using standard statistical notation and nomenclature, I'll present the basic principles of language modeling, starting from the Markovian n-gram language models of Shannon (1948) and connections to compression. Then I will fast forward 70 years and present the modern transformer architecture introduced by Google (2017). I'll conclude with a discussion of the scale at which OpenAI has been able to fit foundational large language models and apply reinforcement learning with human feedback fine tuning to align them with the goals of being helpful, truthful, and harmless. |
|
1:45 - 2:30 PM
Large Language Models Beyond Language by Sachit Menon Department of Computer Science Columbia University Abstract The impact of large language models (LLMs) on any tasks involving the creation of text in response to text has become clear in recent months. What has gone less noticed is the potential impact on areas that are not directly language related, such as computer vision and robotics. I will discuss new paradigms of using LLMs as reasoners, especially via program generation, that stand to bring substantial change to fields that are seemingly unrelated to the natural-language origins of LLMs. |
|
2:30 - 3:00 PM
Coffee Break |
|
3:00 - 3:45 PM
Evaluating the (Moral) Beliefs Encoded in LLMs by Claudia Shi Department of Computer Science Columbia University Abstract This talk focuses on large-scale automatic evaluations of LLMs, covering design principles, evaluation protocols, and a case study analyzing the moral beliefs of 28 LLMs. |
|
3:45 - 4:30 PM
The Evolving World of Natural Language Processing in Healthcare and Life Sciences Presentation Slides https://www.nycasa.org/ASA-LLM-Workshop-Jul-24-2023-IQVIA-Slides.pdf by Marjan Kamyab IQVIA NLP and Kaitlyn Whyte IQVIA NLP Abstract In this presentation from IQVIA, our speakers will cover the trends in NLP, and how this is impacting the way in which they are leading and delivering complex NLP projects across the healthcare and life sciences industries. As part of the Linguamatics team, a company acquired by IQVIA in 2019, the team has been building and delivering NLP solutions to these markets for over 20 years, and so has a breadth and depth of experience in applying the different techniques that have emerged over that time in the field. This talk will focus on choosing the right task for the job at hand, taking into account three key axes of decision making - cost, convenience and accuracy. This talk will cover the use of different techniques, from statistical and LLM approaches to linguistic and rules-based solutions, including presenting real world case studies from academia and industry. |
|
4:30 - 5:00 PM
Group Discussion Moderator David Banks Duke University |