[hobs]

You cant prevent an LLM from having a hallucination.

The current state of the art is retrieval + summarization, all that knowledge it was trained on still exists. When performing a search having no reference for the knowledge is a decent signal that it may not exist at all and you may be talking to a liar.

[capableweb]

You can though, by designing your prompt against it and using very low temperature values so it's more deterministic.