Hallucinations

Hallucinations are types of errors Large Language Models make that can include factual inaccuracies and reasoning failures.

The output usually sounds plausible and confident, but isn't grounded in the input or in fact. For example, a model might invent a citation, a quote or an API method that doesn't exist.

They're one of the key challenges for LLM agents that need to plan, discussed in Understanding the planning of LLM agents: A survey.