Generative AI refers to artificial intelligence systems capable of producing content such as text, images, audio, or code. Unlike traditional software that follows explicit instructions, generative AI models are trained on large datasets to learn patterns and generate human-like output. Examples include ChatGPT and Google Gemini. UC Berkeley has several Licensed Generative AI Tools.
Generative AI tools use large language models (LLMs), which are trained on massive text corpora. These models learn statistical patterns in language and can generate coherent text responses based on prompts (i.e. they sound like natural language). ChatGPT and Google Gemini use a type of neural network called a transformer and can make predictions based on prior context.
Key questions include:
AI models reflect the biases of the data used to train them, including stereotypes, biased language, exclusionary, and limited responses. Much of the initial training data was taken, without permission or agreement, from scholars and artists. For these reasons, it is important to approach AI-generated content critically and to remain aware of its potential social implications.
Generative AI can "hallucinate," meaning it produces plausible-sounding but factually incorrect or entirely fabricated information. It does not know facts; it predicts patterns, and isn’t explicitly trained to say “I don’t know” in response to queries. Always fact-check outputs, especially when using generative AI tools for academic work.