BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//euroscipy-2026//talk//H9UDLT
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-euroscipy-2026-H9UDLT@pretalx.com
DTSTART;TZID=CET:20260723T090000
DTEND;TZID=CET:20260723T103000
DESCRIPTION:Large language models (LLMs) have become central to modern scie
 ntific computing\, yet for most practitioners they remain opaque systems -
  input goes in\, text comes out\, and the internal mechanism is a mystery.
  Mechanistic interpretability (MI) is the emerging discipline of reverse-e
 ngineering what specific components of a neural network actually *do*.\nUs
 ing Andrej Karpathy's `microgpt` - a fully self-contained\, 200-line\, dep
 endency-free GPT implementation in pure Python - as our subject\, we syste
 matically dissect what a trained language model has learned. No PyTorch\, 
 no specialised ML frameworks: just the familiar tools applied to a genuine
 ly novel problem.\n\nThe model is tiny by design: 4\,192 parameters\, a 27
 -token vocabulary (a–z + a special token)\, trained on 32\,000 names in 
 roughly one minute on a laptop. This makes it the ideal subject for interp
 retability work - every attention weight is inspectable\, every embedding 
 printable\, every head ablatable. The scientific question driving the tuto
 rial is: *"What has this model actually learned about the structure of nam
 es?"*
DTSTAMP:20260603T195601Z
LOCATION:Room 1.38 (Ground Floor\, Turing)
SUMMARY:A Hands-On Introduction to Mechanistic Interpretability - Vasu Shar
 ma
URL:https://pretalx.com/euroscipy-2026/talk/H9UDLT/
END:VEVENT
END:VCALENDAR