Скачать или смотреть Yejin Choi | The enigma of LLMs: On creativity, pluralism, and paradoxes

Our weekly SRI Seminar Series welcomes Yejin Choi, incoming professor in the Department of Computer Science at Stanford University and Senior Fellow at the Stanford Institute for Human-Centered Artificial Intelligence. Previously, Choi was a Wissner-Slivka Professor and MacArthur Fellow at the Paul G. Allen School of Computer Science & Engineering at the University of Washington. Choi has received numerous accolades for her work, and was recently named an AI2050 Senior Fellow by Schmidt Sciences.

Choi’s research investigates whether AI systems can learn commonsense knowledge and reasoning, moral reasoning, and various other problems in natural language processing, artificial intelligence, and computer vision, including neuro-symbolic integration, language grounding with vision and interactions, and AI for social good. Choi’s 2023 TED Talk, “Why AI is incredibly smart and shockingly stupid,” has been viewed more than 2.5 million times, and she was recently named an AI2050 Senior Fellow by Schmidt Sciences.

Moderator: Sheila McIlraith

Talk title:
“The enigma of LLMs: On creativity, pluralism, and paradoxes”

Abstract:
As artificial intelligence (AI) becomes more intelligent, it has become less intelligible to us, defying our expectations about its capabilities and limitations. Moreover, while AI becomes more human-like in many ways, there remains something deeply inhuman about it. In this talk, I will present recent studies from my group examining the nature of surface-level creativity in LLMs compared to human writers, and exploring the challenges and potential paths toward pluralistic AI alignment. I will then present the Generative AI Paradox hypothesis: for AI systems, at least in their current form, generation capabilities may often exceed understanding capabilities—in stark contrast to human intelligence, where generation (of novels, paintings, etc.) can be substantially harder than understanding.

Suggested reading:
X. Lu, M. Sclar, S. Hallinan, N. Mireshghallah, J. Liu, S. Han, A. Ettinger, L. Jiang, K. Chandu, N. Dziri, Y. Choi, “AI as Humanity's Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text,” (2024).

T. Sorensen, J. Moore, J. Fisher, M. Gordon, N. Mireshghallah, C. M. Rytting, A. Ye, L. Jiang, X. Lu, N. Dziri, T. Althoff, Y. Choi, “A Roadmap to Pluralistic Aligment,” arXiv (2024).

P. West, X. Lu, N. Dziri, F. Brahman, L. Li, J.D. Hwang, L. Jiang, J. Fisher, A. Ravichander, K. Chandu, B. Newman, P.W. Koh, A. Ettinger, Y. Choi, “The Generative AI Paradox: "What It Can Create, It May Not Understand,” arXiv (2023).

About Yejin Choi:
Yejin Choi is the Dieter Schwarz Foundation Professor and Senior Fellow at the Department of Computer Science at Stanford University and the Stanford Institute for Human-Centered Artificial Intelligence (HAI), and was previously Wissner-Slivka Chair in the Paul G. Allen School of Computer Science & Engineering at the University of Washington. Choi is a MacArthur Fellow (class of 2022), named among TIME’s 100 Most Influential People in AI in 2023, is a co-recipient of two Test-of-Time Awards and eight Best and Outstanding Paper Awards at top AI conferences including ACL, ICML, NeurIPS, ICCV, CVPR, and AAAI, as well as the Borg Early Career Award (BECA) in 2018, the inaugural Alexa Prize Challenge in 2017, and IEEE AI’s 10 to Watch in 2016. Choi was a main stage speaker at TED 2023, and a keynote speaker for a dozen conferences across several AI disciplines including ACL, CVPR, ICLR, MLSys, VLDB, WebConf, and AAAI. In 2024, Choi was named a Schmidt Sciences AI2050 Senior Fellow.

Choi’s research interests include fundamental limits and capabilities of large language models, alternative training recipes for small language models, neuro-symbolic integration, commonsense knowledge and reasoning, moral norms and values, pluralistic alignment, and AI safety.

About the SRI Seminar Series:
The SRI Seminar Series brings together the Schwartz Reisman community and beyond for a robust exchange of ideas that advance scholarship at the intersection of technology and society. Seminars are led by a leading or emerging scholar and feature extensive discussion.

Each week, a featured speaker will present for 45 minutes, followed by an open discussion. Registered attendees will be emailed a Zoom link before the event begins. The event will be recorded and posted online.