What's the advantage of doing this, versus becoming good at context management and RAG? I always found trained knowledge unreliable, given that it is lossy by construction.
Do you find all your own (human) trained knowledge unreliable?
Do you find all your own (human) trained knowledge unreliable?