Jailbreak
One word, taught differently in 2 fields. Each use is given apart.
In Theory & Algos
Techniques used to bypass AI safety guardrails and access restricted capabilities.
In AI & ML
An attack using optimized text suffixes or adversarial images to bypass safety filters in aligned language models.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
In Theory & Algos
1In AI & ML
1Connections
5The lectures that define it or assume it, and the concepts next to it.