BEGIN:VCALENDAR
VERSION:2.0
PRODID:ILLC Website
X-WR-TIMEZONE:Europe/Amsterdam
BEGIN:VTIMEZONE
TZID:Europe/Amsterdam
X-LIC-LOCATION:Europe/Amsterdam
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:/NewsandEvents/Archives/2024/newsitem/14760/25
 -April-2024-Computational-Linguistics-Seminar-Paul
 -Röttger
DTSTAMP:20240418T144043
SUMMARY:Computational Linguistics Seminar, Paul Rö
 ttger
ATTENDEE;ROLE=Speaker:Paul Röttger (Università Boc
 coni)
DTSTART;TZID=Europe/Amsterdam:20240425T160000
LOCATION:Room L3.33 at LAB42, Amsterdam Science Pa
 rk 900, plus live streaming on Zoom
DESCRIPTION:Much recent work seeks to evaluate val
 ues and opinions in large language models (LLMs), 
 motivated by concerns around real-world LLM applic
 ations. For example, politically-biased LLMs may s
 ubtly influence society when they are used by mill
 ions of people. Such real-world concerns, however,
  stand in stark contrast to the artificiality of c
 urrent evaluations using multiple-choice surveys a
 nd questionnaires: real users do not ask LLMs surv
 ey questions. In my talk, I will present recent wo
 rk in which we challenge the prevailing constraine
 d evaluation paradigm for values and opinions in L
 LMs. I will also outline the steps we are now taki
 ng to build more realistic unconstrained evaluatio
 ns for political values and opinions in LLMs.
X-ALT-DESC;FMTTYPE=text/html:\n  <p>Much recent wo
 rk seeks to evaluate values and opinions in large 
 language models (LLMs), motivated by concerns arou
 nd real-world LLM applications. For example, polit
 ically-biased LLMs may subtly influence society wh
 en they are used by millions of people. Such real-
 world concerns, however, stand in stark contrast t
 o the artificiality of current evaluations using m
 ultiple-choice surveys and questionnaires: real us
 ers do not ask LLMs survey questions. In my talk, 
 I will present <a href="https://arxiv.org/abs/2402
 .16786" target="_blank" rel="noopener">recent work
 </a> in which we challenge the prevailing constrai
 ned evaluation paradigm for values and opinions in
  LLMs. I will also outline the steps we are now ta
 king to build more realistic unconstrained evaluat
 ions for political values and opinions in LLMs.</p
 >\n
URL:https://projects.illc.uva.nl/LaCo/CLS/
END:VEVENT
END:VCALENDAR