BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:Asia/Kolkata
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
UID:167@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20251215T110000
DTEND;TZID=Asia/Kolkata:20251215T120000
DTSTAMP:20251210T072542Z
URL:https://cds.iisc.ac.in/events/cds-kiac-seminar-cds-102-15th-decemberre
 aders-prefer-outputs-of-ai-trained-on-copyrighted-books-over-expert-human-
 writers/
SUMMARY:CDS-KIAC {Seminar} @ CDS: #102: 15th\, December"Readers Prefer Outp
 uts of AI Trained on Copyrighted Books over Expert Human Writers"
DESCRIPTION:We welcome you to CDS-KIAC talk on 15th December 2025 (Monda
 y). The details are as below:\n\n\n\nSpeaker: Dr. Tuhin Chakrabarty\, Sto
 ny Brook University\, US\nTitle: Readers Prefer Outputs of AI Trained on 
 Copyrighted Books over Expert Human Writers\nDate and Time: December 15\,
  2025: 11:00 AM\nVenue: The Seminar will be held on HYBRID Mode\n# 102 C
 DS Seminar Hall /MICROSOFT TEAMS\n\nPlease click on the following link to 
 join the Seminar:\nMeeting link\n\n\n\nAbstract: \nThe use of copyrighted
  books for training AI models has led to numerous lawsuits from authors co
 ncerned about AI's ability to generate derivative content. Yet it's unclea
 r whether these models can generate high-quality literary text while emula
 ting authors' styles and voices. To answer this\, we conducted a preregist
 ered study comparing MFA-trained expert writers with three frontier AI mod
 els (ChatGPT\, Claude\, and Gemini) in writing up to 450-word excerpts emu
 lating the diverse styles of 50 award-winning authors. Based on blind pair
 wise evaluations by 159 representative expert and lay readers\, we found t
 hat when AI was simply prompted to imitate an author\, experts overwhelmin
 gly preferred human writing—both for capturing the author's style and fo
 r overall quality. But when ChatGPT is finetuned on an author's complete w
 orks\, these results flipped dramatically: experts now preferred the AI-ge
 nerated text on both measures. Lay readers showed similar shifts. Fine-tun
 ing eliminates detectable AI stylistic quirks like overuse of clichés tha
 t readers find off-putting in standard AI output\, which largely explains 
 this reversal. These fine-tuned outputs also fooled AI detectors 97% of th
 e time\, in addition to being 99.7% cheaper compared to typical profession
 al writer compensation. Together\, these results have direct implications 
 for ongoing copyright lawsuits\, specifically the legal question of whethe
 r AI training harms the market for authors' work.\n\nBio of Speaker:\nTuhi
 n Chakrabarty is Assistant Professor in the Computer Science Department at
  Stony Brook University (SUNY). Prior to this\, he obtained his PhD from C
 olumbia University. His research interests are broadly in AI\, NLP\, and H
 uman-AI Interaction\, and his work often relies on knowledge\, methods\, a
 nd perspectives from multiple disciplines to address complex problems that
  cannot be fully understood or solved within the boundaries of Computer Sc
 ience. Tuhin's work has been covered in MIT Tech Review\, Bloomberg\, and 
 the Washington Post\, and he has received a Best Paper Honorable Mention a
 ward at ACM CHI and an Outstanding Position Paper award at ICML. Most rece
 ntly\, his research on Generative AI and Fair Use is being profiled by The
  New Yorker\, and his empirical findings on market dilution and substituti
 on effects are being used by leading U.S. law firms in ongoing AI copyrigh
 t litigation.\n\nAbout: This talk is part of the KIAC Seminar Series 
 on AI for Fintech.\n\nHost Faculty: Danish Pruthi\, CDS\n\n\n\nALL ARE WE
 LCOME
CATEGORIES:Events,Talks
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20241215T110000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR