ChatGPT is awesome but on its own it does not help much with enterprise search. First, ChatGPT does not know anything about your data. Second, even if it did (think fine-tuning), it would be prone to providing incorrect answers (hallucinations) and often unable to link its answers back to the sources of information. The approach that overcomes these difficulties is called Retrieval Augmented Generation (RAG). RAG consists of two components: a "retriever" that finds information relevant to the question and a large language model (e.g., ChatGPT) that answers the question based on the information provided by the retriever. In addition to that, enterprise search has other dimensions that need to be implemented in a RAG system:
— permissions: users search only over the content they are permissioned to see
— "faceted" search where you can filter / prioritize by categories configured based on your data such as the data source (e.g., knowledge base articles may take precedence over tickets), the freshness of the document (when was it last modified), the author of the document, etc
— ability to learn from feedback both implicit and explicit (e.g., upvote/downvote documents and specify the right document if it's not found)
More details about the components of Serenity are in our whitepaper above.