Do I need to define collate? Maybe it wasn’t the best choice of verbiage but the point still stands. The quality of the output is always relative to the input. That’s why a growing number of companies are training their own llms with data from their own databases instead of trying to rely on external datasets.

For the record, I’m not talking about ones that you can ask a question and get an answer. I was talking about law firms using a local or privately hosted llm to scan through discovery documents and finding keywords or related keywords that may be relevant to the case they are working. Especially now that a lot of discovery is digital.

I can’t give more detail than the following because it may not be public yet but I am aware of one company working on their own llm to let clients more easily find info that has been published on their platform and would take longer to skim through than to just as a search engine.