BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Can Language Models Learn Truthfulness? - He He\, New York Univers
 ity
DTSTART:20240118T150000Z
DTEND:20240118T160000Z
UID:TALK210634@talks.cam.ac.uk
CONTACT:Panagiotis Fytas
DESCRIPTION:Today's large language models (LLMs) are trained on vast amoun
 ts of text from the internet\, which contains both factual and misleading 
 information about the world. Can language models discern truth from falseh
 ood in this contradicting data? This talk introduces a hypothesis for how 
 LLMs can model truthfulness. Inspired by the agent-model view of language 
 models\, we hypothesize that they can cluster truthful text by modeling a 
 truthful persona: a group of agents that are likely to produce truthful te
 xt and share similar features. I will discuss both results on real data an
 d controlled experiments on synthetic data that support the hypothesis. Ov
 erall\, our findings suggest that models can exploit hierarchical structur
 es in the data to learn abstract concepts like truthfulness.
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR