Generative AI
Generative AI creates new content: text, images, audio, code. It learns a distribution, then samples. LLMs predict the next token. Image models denoise or decode latents. Useful copilots. Risks: hallucination, copyright, deepfakes, leaking training data.
Not the same as a classifier that only outputs spam/not-spam. Generative = produce new samples. Discriminative = label existing ones. Say that pair.
Generative AI — output — three lines. Generative creates samples; still can hallucinate.
Generative vs discriminative + one risk.