LLMs obviously feature extremely advanced natural language understanding, but for many many applications, you can't just ask the LLM your question and get out the answer. The knowledge will often be missing, or the scale and latency requirements would be infeasible.

→ View original post on X — @honnibal