BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Student Spotlight: Yan Xia\, James Ackland\, and Nikolay Petrov - 
 Yan Xia (Aalto University)\, James Ackland (Cambridge)\, and Nikolay Petro
 v (Cambridge)
DTSTART:20240306T150000Z
DTEND:20240306T160000Z
UID:TALK212200@talks.cam.ac.uk
CONTACT:Yara Kyrychenko
DESCRIPTION:This talk is open to the general public.\n\nMeeting ID: 329 28
 7 585 675 \n\nPasscode: yKwfhf\n\n\nYan Xia (Aalto University):\n\nTitle: 
 Integrated or Segregated? User Behavior Change after Cross-Party Interacti
 ons on Reddit\n\nAbstract:\nIt is a widely shared concern that social medi
 a reinforces echo chambers of like-minded users and exacerbates political 
 polarization. While fostering interactions across party lines is recognize
 d as an important strategy to break echo chambers\, there is a lack of emp
 irical evidence on whether users will actually become more integrated or i
 nstead more segregated following such interactions on real social media pl
 atforms. We fill this gap by inspecting how users change their community p
 articipation after receiving a cross-party reply in the U.S. politics disc
 ussion on Reddit. More specifically\, we investigate if their participatio
 n increases in communities of the opposing party\, or in communities of th
 eir own party. We find that receiving a reply is significantly associated 
 with increased user activity in both types of communities\; when the reply
  is a cross-party one\, the activity boost in cross-party communities is w
 eaker. Nevertheless\, compared with the case of receiving no reply\, users
  are still significantly more likely to increase their participation in cr
 oss-party communities after receiving a cross-party reply. Our results the
 refore hint at a depolarization effect of cross-party interactions that be
 tter integrate users into discussions of the opposing side.\n\n\nJames Ack
 land (Cambridge):\n\nTitle: The Geographical Psychology of Ideological Mis
 alignment\n\nAbstract:\nPolitical psychologists have debated whether ideol
 ogy is constructed from the top-down\, by national-level parties and elite
 s forming packages of beliefs to "sell" to voters (Downs\, 1957)\; or from
  the bottom-up\, by voters themselves aligning policy preferences with mor
 e fundamental social and psychological needs (Duckitt & Sibley\, 2010). In
  this work\, I assume that both processes coexist\, and show how their int
 eraction can explain some of the phenomena that characterise our modern po
 litics. Of particular interest are places where bottom-up preferences are 
 not matched by the top-down political offering. In Western Europe\, this o
 ften means places where social conservatism exists alongside left-leaning 
 economic preferences\, in contrast to the pairing of social conservatism w
 ith a free-market ideology at the national level. In such places\, I hypot
 hesise that populist politics will be more successful\, as measured by vot
 ing behaviour and political attitudes.\n\nNikolay Petrov (Cambridge):\n\nT
 itle: Limited ability of LLMs to simulate human psychological behaviours: 
 an in-depth psychometric analysis\n\nAbstract:\nThe humanlike responses of
  Large Language Models (LLMs) have prompted social scientists to investiga
 te whether LLMs can be used to simulate human participants in experiments\
 , opinion polls and surveys. Of central interest in this line of research 
 has been mapping out the psychological profile of LLMs by prompting them t
 o respond to standardized questionnaires. The conflicting findings of this
  research are unsurprising given that going from LLMs' text responses on s
 urveys to mapping out underlying\, or latent\, traits is no easy task. To 
 address this\, we use psychometrics\, the science of psychological measure
 ment. In this study\, we prompt OpenAI's flagship models\, GPT-3.5 and GPT
 -4\, by asking them to assume different personas and respond to a range of
  standardized measures of personality constructs. We used two kinds of per
 sona descriptions: either generic (5 random person descriptions) or specif
 ic (mostly demographics of actual humans from a large-scale human dataset)
 . We found that using generic persona descriptions\, more powerful models\
 , such as GPT-4\, show promising abilities to respond coherently\, and sim
 ilar to human norms\, but both models failed miserably in assuming specifi
 c personas\, described using demographic variables. We conclude that\, cur
 rently\, when LLMs are prompted to simulate specific human(s)\, they canno
 t represent latent traits and thus their responses fail to generalize acro
 ss tasks.
LOCATION:Nick Mackintosch Room\, Department of Psychology\, Downing Site
END:VEVENT
END:VCALENDAR