BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:AI Center Seminar - AI Fundamentals series - Simon Schrodi - "Towa
 rds a Science of AI (Safety)"
DTSTART:20261002T110000
DTEND:20261002T120000
DTSTAMP:20260930T042046Z
UID:a8eeaab413d3083f8ed452f2cb0df835cdb58a3e87e97dbc61fe9686
CATEGORIES:Conferences - Seminars
DESCRIPTION:Simon Schrodi\nThe talk is jointly organized by the EPFL AI 
 Center and the DLAB as part of the AI fundamentals seminar series.\n\nH
 ost: Raghav Singhal\n\nTitle\nTowards a Science of AI (Safety)\n\nAbstract
 \nAI models have remarkable capabilities\, yet their successes and failure
 s often remain poorly understood. More concerningly\, what they learn can 
 differ from what we intended. Understanding these gaps matters for assessi
 ng their reliability and safety. In this talk\, I show how controlled expe
 rimentation and mechanistic analysis can help explain how models learn and
  generalize. Using examples from my research\, I show that certain data pr
 operties but also internal mechanisms shape generalization\, and how model
 s can learn behaviors not explicitly expressed in their fine-tuning data. 
 Together\, these examples are steps towards a science of AI that helps us 
 understand when\, how\, and why model behave as they do\, providing us wit
 h a stronger foundation for their safe deployment.\n\nBio\nSimon Schrodi i
 s a fifth-year PhD student at the University of Freiburg\, advised by Prof
 . Thomas Brox. His research lies at the intersection of the science of dee
 p learning\, interpretability\, and AI safety. In particular\, he investig
 ates how AI models learn and generalize\, and how this shapes their behavi
 or using controlled experiments and mechanistic analysis. His broader inte
 rests also include machine learning for climate science and automated mac
 hine learning. Simon is also a research scholar at MATS 9\, mentored by A
 lex Cloud\, Cem Anil\, and Arthur Conmy. His work there examines the limit
 ations of pretraining safety methods and how pretraining shapes models' pr
 opensity for agentic misalignment. Previously\, he completed his MSc in co
 mputer science at the University of Freiburg in 2022\, following BEng stud
 ies at the Cooperative State University Karlsruhe. More information can be
  found at https://simonschrodi.github.io.\n 
LOCATION:ELE 117 https://plan.epfl.ch/?room==ELE%20117 https://epfl.zoom.u
 s/j/65340485143?pwd=2QsclxuwH4fFL3Bb0Y9wZzFuWGkdbL.1
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR