Comment by nico

Interesting

I think soon we are going to realize that we don’t really need training the models

We just need good indexing and sampling

Essentially at some level any LLM is equivalent to a DB of the dataset, with a great NLP interface on top

Both are just different methods of navigating stored data