BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//speaker//38ZQA8
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-pydata-amsterdam2026-NPZNKA@pretalx.com
DTSTART;TZID=CET:20260910T133000
DTEND;TZID=CET:20260910T141500
DESCRIPTION:As Large Language Models (LLMs) transition into core enterprise
  infrastructure\, treating them as "black boxes" is no longer an acceptabl
 e engineering risk. Can we turn LLM agents into debuggable\, predictable s
 oftware systems?\n\nThe answer lies in an accelerating field of research m
 ade famous by an LLM [obsessed with the Golden Gate Bridge](https://www.an
 thropic.com/news/golden-gate-claude): **mechanistic interpretability (mech
 interp)**. This discipline aims to _reverse-engineer neural networks_ into
  human-understandable algorithms.\n\nAimed at data scientists and AI engin
 eers familiar with transformer basics\, this talk showcases what the curre
 nt state of the art can actually achieve. Moving past theoretical toy exam
 ples\, we will explore recent breakthroughs in extracting high-level seman
 tic features\, such as **isolating abstract concepts like emotional subtex
 t** within model weights.\n\nYou will learn the intuitions behind how we r
 econstruct model features from LLM layers\, how they can be **actively man
 ipulated via activation steering** to change model behaviour without fine-
 tuning\, and a realistic assessment of where current tooling hits a hard w
 all.
DTSTAMP:20260710T150510Z
LOCATION:Unconference
SUMMARY:Inside the Mind of an LLM - Luca Baggi
URL:https://pretalx.com/pydata-amsterdam2026/talk/NPZNKA/
END:VEVENT
END:VCALENDAR