[Theory] TOMORROW: 3/24 TTIC Colloquium: Eunsol Choi, New York University

[Mary Marre via Theory]

*When:*        Monday, March 24, 2025 at* 11:30** am CT *


*Where:       *Talk will be given *live, in-person* at

                   TTIC, 6045 S. Kenwood Avenue

                   5th Floor, Room 530


*Virtually:*    *livestream via panopto
<https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=cc5bda68-8942-4724-87f3-b2a7017990db>*

                 * Access restricted to TTIC/UChicago**





*Who: *         Eunsol Choi, New York University



Title:          Equipping LLMs for Complex Knowledge Scenarios: Retrieval
and Interaction

Abstract: Natural language provides an intuitive and powerful interface to
access knowledge at scale. However, language models are largely optimized
for autoregressive response generation and not as an information access
interface. In this talk, I will present two key ingredients for more robust
information interfaces: retrieval augmentation and multi-turn interaction.
In the first part of the talk, I will discuss retrieval augmentation. While
being touted for reducing hallucination, it poses new challenges such as
scalability and cascading errors when retrieval fails. To address this, we
learn a compressor that maps retrieved documents into textual summaries
prior to in-context integration. This process not only reduces the
computational cost but also filters irrelevant or incorrect information. In
the second part of the talk, I will discuss the challenges of ambiguous and
underspecified user inputs. LLMs are almost universally optimized for
providing the best single-turn answer to a query, disregarding their
potential as dialogue agents. We propose a system that can interact with
users to clarify their intent before answering. By simulating their
expected outcomes in future turns, we reward LMs for generating clarifying
questions and not just answering immediately. Together, the talk outlines
key research directions for transforming LMs into an interface to answer
information seeking questions.

Bio: Eunsol Choi is an assistant professor of Computer Science (Courant
Institute) and Data Science at New York University. Her research spans
natural language processing and machine learning, with a focus on
interpreting and reasoning about text in dynamic real-world contexts. Prior
to joining NYU, she was an assistant professor at the University of Texas
at Austin. She also spent a year at Google AI as a visiting researcher. She
holds a Ph.D. in computer science and engineering from the University of
Washington. She is a recipient of a Facebook research fellowship, Google
faculty research award, Sony faculty award, and an outstanding paper award
at EMNLP.

*Host: **Karen Livescu* <klivescu at ttic.edu>

*Access to this livestream is limited to TTIC / UChicago (press panopto
link and sign in to your UChicago account with CNetID).


Mary C. Marre
Faculty Administrative Support
*Toyota Technological Institute*
*6045 S. Kenwood Avenue, Rm 517*
*Chicago, IL  60637*
*773-834-1757*
*mmarre at ttic.edu <mmarre at ttic.edu>*


On Tue, Mar 18, 2025 at 10:32 AM Mary Marre <mmarre at ttic.edu> wrote:

> *When:*        Monday, March 24, 2025 at* 11:30** am CT *
>
>
> *Where:       *Talk will be given *live, in-person* at
>
>                    TTIC, 6045 S. Kenwood Avenue
>
>                    5th Floor, Room 530
>
>
> *Virtually:*    *tba*
>
>
>
>
>
> *Who: *         Eunsol Choi, New York University
>
>
>
> Title:          Equipping LLMs for Complex Knowledge Scenarios: Retrieval
> and Interaction
>
> Abstract: Natural language provides an intuitive and powerful interface
> to access knowledge at scale. However, language models are largely
> optimized for autoregressive response generation and not as an information
> access interface. In this talk, I will present two key ingredients for more
> robust information interfaces: retrieval augmentation and multi-turn
> interaction. In the first part of the talk, I will discuss retrieval
> augmentation. While being touted for reducing hallucination, it poses new
> challenges such as scalability and cascading errors when retrieval fails.
> To address this, we learn a compressor that maps retrieved documents into
> textual summaries prior to in-context integration. This process not only
> reduces the computational cost but also filters irrelevant or incorrect
> information. In the second part of the talk, I will discuss the challenges
> of ambiguous and underspecified user inputs. LLMs are almost universally
> optimized for providing the best single-turn answer to a query,
> disregarding their potential as dialogue agents. We propose a system that
> can interact with users to clarify their intent before answering. By
> simulating their expected outcomes in future turns, we reward LMs for
> generating clarifying questions and not just answering immediately.
> Together, the talk outlines key research directions for transforming LMs
> into an interface to answer information seeking questions.
>
> Bio: Eunsol Choi is an assistant professor of Computer Science (Courant
> Institute) and Data Science at New York University. Her research spans
> natural language processing and machine learning, with a focus on
> interpreting and reasoning about text in dynamic real-world contexts. Prior
> to joining NYU, she was an assistant professor at the University of Texas
> at Austin. She also spent a year at Google AI as a visiting researcher. She
> holds a Ph.D. in computer science and engineering from the University of
> Washington. She is a recipient of a Facebook research fellowship, Google
> faculty research award, Sony faculty award, and an outstanding paper award
> at EMNLP.
>
> *Host: **Karen Livescu* <klivescu at ttic.edu>
>
>
>
> Mary C. Marre
> Faculty Administrative Support
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue, Rm 517*
> *Chicago, IL  60637*
> *773-834-1757*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20250323/e9652063/attachment-0001.html>