Generative AI creates new content by learning patterns in existing data. Earlier generative models were used chiefly to explore numerical data; deep learning extended the approach to text, images, audio, and video. An output can look realistic without being reliable. section 5.1 section 5.12
Practical uses include:
- Drafting text and generating images or video so creators can spend more time refining ideas. section 5.7
- Answering routine customer-support questions using software documentation. section 5.7
- Producing code suggestions and summaries of information, which still need checking. section 5.7 section 5.12
A convincing answer can be false
A lawyer used ChatGPT for legal research and submitted a brief citing six nonexistent cases. When asked whether one case was real, ChatGPT affirmed that it was. Coherent wording and confident answers are not substitutes for checking sources. section 5.10
Output quality depends on input data and prompts. Generative systems can fabricate facts, produce harmful content, or be manipulated; their use also raises questions about rights to the creative work used in training. Treat generated work as a draft or aid, especially where accuracy, safety, or ownership matters. section 5.12 section 5.10 section 5.9
