BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:EPFL AI Center x UK AI Security Institute - mini-symposium — Eva
 luations\, Impact\, and the Science of AI Safety
DTSTART:20250929T143000
DTEND:20250929T200000
DTSTAMP:20260429T004226Z
UID:21b8e83201139a6b046b3ee5b0c1104fd24a3b598c3ddada0e473ccb
CATEGORIES:Conferences - Seminars
DESCRIPTION:See program \nJoin us at the EPFL AI Center on Monday\, Septe
 mber 29\, for a mini-symposium—“Evaluations\, Impact\, and the Science
  of AI Safety.”\n\nOrganized with the UK AI Security Institute (AISI) a
 nd the EPFL NLP Lab\, the program will dive into AISI’s current prioriti
 es\, methods for sandboxing agentic evaluations\, fresh evidence on the re
 al-world impacts of AI advice\, and the science of evals. We’ll wrap up 
 with a poster session and networking apéro.\n\nBring a poster—the apero
  session will highlight EPFL student work. When registering\, select “B
 ring a poster” and add your title (limited slots).\n\n! Registration is
  required to attend this event. Access is restricted to EPFL participants.
  !\nRegistration form (restricted to EPFL email addresses): https://forms
 .office.com/e/wPBhzPbzFP\n\nProgram\n\n	14:30 – 15:00 — Welcome and ch
 eck-in\n	 \n	15:00 – 15:05 — Welcome\, Prof. Antoine Bosselut\, EPFL
 \n\n\n\n	15:05 – 15:20 — Opening & overview of AISI\, Shannon Yang\, 
 UK AI Safety Institute (DSIT)\n	 \n	15:20 – 15:40 — Sandboxing agenti
 c evaluations\, Art O’Cathain\, AISI\n	 \n	15:40 – 16:00 — “AI u
 se in political information search”\, by Lennart Luettgau\n	 \n	16:00 
 – 16:15 — Short break\n	 \n	16:15 – 16:35 — The science of evalua
 tions\, Magda Dubois\, AISI\n	 \n	16:35 – 17:00 — "Fundamental Limi
 tations in Defending LLM Finetuning APIs" by Eric Winsor\n\n17:00 — Aper
 o & Posters from EPFL students\n\nTopics covered\n\n\n	Agentic evaluation
 s & sandboxing frameworks\n	Measuring real-world impacts of AI advice & we
 llbeing\n	Science of evals: robust\, transparent benchmarking\n	Trustworth
 y AI practices: reproducibility\, transparency\, ethics\n\nRegistration is
  free but mandatory.\n\n 
LOCATION:ELE 117 https://plan.epfl.ch/?room==ELE%20117
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR
