BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Understanding LLMs via their Generative Successes and Shortcomings
 . - Swabha Swayamdipta\, University of Southern California
DTSTART:20240208T160000Z
DTEND:20240208T170000Z
UID:TALK212032@talks.cam.ac.uk
CONTACT:Panagiotis Fytas
DESCRIPTION:Generative capabilities of large language models have grown be
 yond the wildest imagination of the broader AI research community\, leadin
 g many to speculate whether these successes may be attributed to the train
 ing data or different factors concerning the model. I will present some wo
 rk from my group which has revealed unique successes and shortcomings in t
 he generative capabilities of LLMs\, on knowledge-oriented tasks\, tasks w
 ith human and social utility and tasks that reveal more than surface-level
  understanding of language. I will also discuss some aspects of language g
 eneration itself and why algorithms like truncation sampling have been so 
 successful.
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR