I mean... That's definitely a "why not both" situation. 1. make the (compressed) Wikipedia searchable better as a knowledge base 2. use the LLM as a "interface

I mean... That's definitely a "why not both" situation. 1. make the (compressed) Wikipedia searchable better as a knowledge base 2. use the LLM as a "interface" to that knowledge base I investigated 1. back when all of (English, text-only) Wikipedia was about 2 GB. Maybe it is time to look at that toy code base again.