Hacking AI - How to Survive the AI Uprising
summary
Shows concrete attacks and defenses: early white‑text resume tricks fail but simple hidden tokens like greedy coordinate gradients and the Open Sesame universal blackbox string jail break LLMs (claims ~100% on known models and ~85–97% across many models), teachers catch students with trigger words Frankenstein and goober, image defenses Nightshade and Glaze intentionally poison or explode gradients, inaudible dolphin audio activates devices, three roadside stickers fool a Tesla, and Fox plus NicorNot.com demonstrate practical facial‑recognition evasion that actually works.
AI-written summary · about
chapters (7)
- 0:00Welcome to the resistance
- 1:09Hacking AI in 2024
- 6:56Text hacking with LLMs
- 17:16AI data poisoning attacks
- 26:27AI vision adversarial tricks
- 37:12Ethics and AI control
- 42:09Closing and call to action
related talks
37mKeynote - Hacking AI How to Survive the AI Uprising
35mHacking AI Systems: How to (Still) Trick Artificial Intelligence
42mHow to attack AI systems (and how to defend them) !!!!
46mA Short Introduction to Hacking Artificial Intelligence
37mHow to Break AI Systems (Before Someone Else Does)
45mSecuring AI: There Is No Try, Only Do!
53mHow to Break AI Systems (Before Someone Else Does)
52mSleeping With One AI Open: An Introduction to Attacks Against...