I'm a bit confused about the terminology. Is Generative AI just a new name for Machine Learning, or is there a fundamental difference in how these models work? If I'm already familiar with basic ML concepts like regression and classification, what new things do I need to learn to understand GenAI?
3 answers
Think of it this way: Traditional ML is "Discriminative." It looks at data and classifies it (e.g., "Is this a cat or a dog?"). Generative AI, on the other hand, creates new data (e.g., "Draw me a picture of a cat in a tuxedo"). Under the hood, GenAI uses architectures like Transformers or GANs (Generative Adversarial Networks). Since you know ML, you'll find the concepts of weights and biases familiar, but you'll need to dive into "Attention Mechanisms" and how models predict the next token in a sequence.
Heather, since you mentioned Transformers, do you think they have completely replaced GANs for image generation now, or do GANs still have a place in the industry?
The main shift is from "predicting a label" to "predicting a distribution." It’s a much more complex way of handling probability!
Spot on, Olivia. It's the difference between picking one answer and being able to generate infinite variations of that answer.
Actually, Charles, GANs are still very relevant for high-speed, real-time generation, like in video games or deepfake detection. However, Diffusion models (which are different from Transformers) have mostly taken over for high-quality static images like what you see in Midjourney. Each has its own strengths depending on whether you need speed or sheer artistic quality.