10-14, 10:30–12:30 (Canada/Eastern), Workshops & Speed
Language: English
Let's get hands on and dive into the marvelous world of "artificial intelligence" and "prompt engineering"!
This session will cover two CTF challenges based on Stable Diffusion which is a model capable of generating photo-realistic images given any text input. This will show in a practical fashion some real impact of something otherwise artificial.
The challenges are going to be solved step by step alongside a dive into diffusion models and technical details on machine learning.
We will provide a dedicated online environment that requires low computing power on your end to do prompt hacking and try out Stable Diffusion.
Prerequisites:
This workshop supports three levels of technical readiness:
1) To participate in prompt hacking, bring any laptop or cellphone with a Web browser
2) For a more low-level challenge, bring a laptop with Python installed
3) To optionally play with Stable Diffusion tools locally, a laptop with dedicated 3D graphic card is required
L'aide aux participants en Français sera aussi disponible.
This workshop uncovers some details about machine learning and models that leads the participants to form a proper yet simple technical understanding of how image related "AI" works.
Outline:
- Intro (15mins)
What is Stable Diffusion
What is a Model and Prompt Engineering
-
Workshop Part 1 (45mins)
Model RCE
Doable with a laptop and Python
This will be classic Python Pickling, requires some programming skills
Solution given after 30mins -
Break (15mins)
Will also be used to catch up with some participants -
Workshop Part 2 (45mins)
Doable with cellphone or laptop Web browser
This will be finding what’s hiding in the deep learning brain
This is bleeding edge prompt hacking and is surprisingly accessible by everyone
Solution given after 30mins -
Bonus: Hypernetworks
If time permits, we can look deeper at Hypernetworks and how they are being trained. This is related to the challenges above as the base of it, but also opens the topic to ethical and privacy concerns.
No
Jonathan is part of NorthSec as a CTF challenge designer. He is passionate about Application Security and enjoys architecture analysis, code review, threat modeling and debunking security tools. Jonathan holds a bachelor's degree in Software Engineering from ETS Montreal and has 20 years of experience in Information Technology and Security.