I've been researching hallucination reduction/elimination morning through night for over a year. The only effective techniques that I have found all involve rewriting the context that is sent to the LLM. The context needs to be structured in a way that prevents the LLM from going off the rails.

For example, coreference resolution significantly reduces hallucinations. I wrote a short article on this here: https://medium.com/@michaelwood33311/creating-accurate-ai-coreference-resolution-with-fastcoref-20f06044bdf9.

However, coreference resolution alone is still insufficient to fully eliminate hallucination. Total elimination requires extensive rewriting of the context such as converting all information to Fully-Formatted Facts: https://www.ragfix.ai/fully-formatted-facts-master-key-to-eliminating-hallucinations/

If you discover other techniques of hallucination mitigation/elimination please let me know! I'm constantly in search of learning more on the topic which lead me to your great article.