Asking a chatbot a question
How a language model responds to a prompt, why wording matters and why the same question may produce different answers.
Large language model case study
What this example shows
Follow one question from the words a person enters to the generated response they read.
What goes in
A written prompt containing the question, instructions and any context the person chooses to provide.
What the AI system does
The service splits the text into tokens. A language model uses learned patterns to repeatedly predict likely next tokens, while the surrounding product may add instructions, tools or retrieved information.
What comes out
A generated response presented as a conversation. It may include text, links or other media depending on the service.
What can go wrong
The response may invent facts, misunderstand the question, omit context or reproduce patterns of bias. Private details in the prompt may be stored or processed by the provider.
Who is responsible
The provider is responsible for how the service is designed and operated. The person using an answer remains responsible for checking important claims before relying on them.