BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pyconde-pydata-2026//talk//3JLSEF
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-pyconde-pydata-2026-3JLSEF@pretalx.com
DTSTART;TZID=CET:20260414T163000
DTEND;TZID=CET:20260414T170000
DESCRIPTION:With Large Language Models (LLMs)\, generating high-quality tex
 t and images is easy and so is\nmisusing it. As AI-generated content becom
 es harder to distinguish from human generated content\,\ndevelopers are in
 creasingly asking: How can we verify whether a piece of text comes from an
  LLM?\nWe’ll explore Python’s simplicity and rich ecosystem of librari
 es to solve this problem.\n\nThis talk introduces the foundations of LLM w
 atermarking and shows how developers can implement\nthese techniques entir
 ely in Python. We’ll discuss two core approaches\, EXP sampling method a
 nd\nKGW method. We will go through the implementation of the KGW method us
 ing simple\,\ntransparent code\, and compare it with the EXP approach. The
 re's no need for a large model or a GPU\ncluster to understand how these s
 ystems work and the core ideas can be implemented in pure\nPython using si
 mple code. The code repositories\, which includes both methods will be pro
 vided so\nthat the attendees can follow along.\n\nAlong the way\, we’ll 
 discuss the trade-offs and the limitations of current research. And for th
 ose\nwondering\, “Do I have to implement all this myself?”\, the talk 
 concludes with a quick overview of MarkLLM\, an existing open-source toolk
 it that provides a unified Python interface for experimenting with waterma
 rking algorithms.\n\nAttendees will leave with a clear understanding of ho
 w watermarking works\, when it’s useful\, and\nhow to integrate these te
 chniques into real-world Python projects.
DTSTAMP:20260523T180014Z
LOCATION:Dynamicum [Ground Floor]
SUMMARY:Catch the LLM if you Can: Watermarking LLMs - Subhosri Basu
URL:https://pretalx.com/pyconde-pydata-2026/talk/3JLSEF/
END:VEVENT
END:VCALENDAR