Generative AI creates new content: text, images, audio, video or code. Earlier "discriminative" AI mostly decided between options (is this email spam?). Generative AI produces something that did not exist before, based on patterns it learned from large amounts of data.
What it can generate
- Text: answers, emails, summaries, resumes, cover letters.
- Images: pictures and edits from a written description.
- Audio and video: speech, music, short clips.
- Code: functions, tests and explanations.
Foundation models and LLMs
A foundation model is a large model trained on broad data that can be adapted to many tasks. A large language model (LLM) is a foundation model for text. A multimodal model works with more than one type of input, such as text plus images or PDFs.
Open and closed models
- Closed models are used through an API or an app (Claude, GPT, Gemini). You do not get the weights.
- Open-weight models can be downloaded and run on your own hardware (for example Llama or Mistral models). You manage hosting, but you control the data and cost.
Strengths and limits
Generative AI is excellent at drafting, transforming and explaining text and at working with natural language. It is weaker at exact facts, current events, arithmetic and anything needing guaranteed correctness. Good products combine the model with your own data, tools and checks.