All videos

AI Alignment is More Fragile Than You Think (And Shaggoth is Lurking)

September 30, 2025

One of the fundamental features of LLMs is their ability to provide responses aligned with human ethical principles and intentions. In his talk, Przemek will explore what LLMs truly "know" about morals, discuss how models are being trained for AI Alignment, and reveal how (surprisingly) easy it is to break. He'll show experimental results, in which a small set of harmful examples could derail alignment. A live demo and open discussion on what it all means for us will follow.

About the speaker

Przemek Maciolek

Przemek Maciolek

Staff Software Engineer
By day, working on Data Engineering at a major Home-sharing Platform. In past lives, worked for numerous startups, where he was a founding engineer, principal…
Read more

Other videos that you might like

fallback image
Reactive systems
A story about connecting microservices
Paweł Kamiński
fallback image
BUGinnings in IT
UI/UX
All you need to know to land and manage your first UX job
Monika Soja, Mira Melhor
fallback image
Sphere.it conf 2022
How I gained from gracing the world with yet another NPM package
Mikołaj Klaman
AI Alignment is More Fragile Than You Think (And Shaggoth is Lurking)
Beyond Senior
Sphere.it Conf 2023
The most important 5 minutes of a new leader
Dawid Ostręga
All videos