For many NLP tasks (which is what I mostly use LLMs for), hallucinations can be prevented with simple, procedural checks against the input or a controlled vocab

For many NLP tasks (which is what I mostly use LLMs for), hallucinations can be prevented with simple, procedural checks against the input or a controlled vocabulary. For example, for NER tasks, you can just check whether the extracted entities are valid relative to either of the two.