Generative AI
Also known as: GenAI, generative artificial intelligence
Traditional AI systems mostly classified or predicted: is this email spam, will this customer churn, what's in this image. Generative AI creates. Give a generative model a prompt and it produces something new: a paragraph, an image, a piece of music, a video clip, a working function. This shift from 'predict existing things' to 'create new things' is what made the current wave of AI feel qualitatively different.
The most common generative AI architectures include transformer-based LLMs (for text and code), diffusion models (for images and video), and hybrid multimodal systems that combine both. The key training insight is that a model that learns to accurately predict missing parts of data, whether the next word in text or a corrupted patch in an image, learns the structure of that data well enough to generate new examples.
Generative AI is the umbrella term that encompasses LLMs, image generators like DALL-E, video generators like Sora and Veo, audio synthesis, and code generation. For builders, the practical distinction that matters is which modality or combination of modalities your product needs, since the capabilities, costs, and underlying models differ significantly across the generative AI landscape.