[sureglymop]

Possibly unrelated but something I never fully understood: while we can't create a perfect parser for natural language, why don't we optimistically parse it to extract semantics and feed that into LLMs as well?