Skip to Main Content

Generative AI and Chatbots

This guide offers advice on generative AI chatbots and tools and how to best use them to support your work.

How LLMs and generative AI work

Generative AI chatbots create new content by predicting the word most statistically likely to appear next. The predictions are based on patterns observed in their training data, which contains incorrect and biased information, and post-training fine tuning using human feedback. Generative AI chatbots are not search engines and do not give a set of results in response to a specific search. They model human language, they do not "know" things.

AI chatbot responses can often include a mix of correct and incorrect information.

Explainer articles

How smart is ChatGPT?

Dr. Stuart J. Russell, a Professor of Computer Science at UC Berkeley, explains Large Language Models with examples from GPT-4.

Duration: <1 minute

Why AI is incredibly smart and shockingly stupid

Computer scientist Yejin Choi explains massive artificial intelligence systems like ChatGPT.

Duration: 16:03 minutes, video highlights:

  • 1:32 Societal level challenges
  • 3:35 AI is passing the bar exam. Does that mean that it has common sense?
  • 9:21 Three types of data that AI is trained on
  • 10:55 Language models do not equal knowledge models