VINCI Seminar featuring Marzena Karpinska
We are pleased to have Marzena Karpinska from Simon Fraser University presenting at our SFU VINCI seminar.

Date: Wednesday, November 25, 2026
Time: 12:30 PM
Location: ASB 10704, Burnaby campus
Please note: This seminar will be held in ASB 10704, not in our usual TASC 1 seminar location.
Zoom Link: Join via Zoom
Talk Title: Authorship identification in the era of large language models
Abstract:
“Who wrote this?” is becoming one of the central questions of the AI era. Modern AI detectors are trained to identify the author behind the prose. However, as AI-assisted writing becomes more common, it matters just as much who contributed the ideas. In the first part of this talk, I will discuss how AI writing differs from human writing at both the prose and content levels, and introduce IdeaLens, an AI detector trained to identify the author of the content rather than the prose.
In the second part, I will extend this question beyond binary classification and show that frontier models are surprisingly good at answering it. Trained on large web corpora containing human writing alongside a growing share of AI-generated text, frontier models effectively become authorship detectors. Given only a single text in a simple prompt, with no web access, tools, or candidate lists, frontier models can answer “who wrote this?” at three increasingly specific levels: (1) whether a human or an AI wrote it, (2) if an AI, which lab created it, and (3) if a human, often exactly who. On the human or AI question, they match trained detectors while providing explanations in natural language. They can also distinguish between major AI labs, and even unmask specific human writers from parametric knowledge alone. I will conclude by showing how exposure in pretraining and reasoning effort affect identifiability, and why limited defenses leave anonymity more fragile than commonly assumed.
Biography:
Marzena Karpinska is an assistant professor at Simon Fraser University in Vancouver, Canada. Her research covers the detection of AI-generated (or translated) content, translation of creative texts, and in broader context human-centred and multilingual natural language processing. Before joining SFU in 2026, she was a senior researcher at Microsoft and a postdoctoral researcher at the University of Massachusetts Amherst. She holds a PhD from the University of Tokyo.