There's a nice critique of the LLM paradigm implicit in approaches from later waves of GOFAI knowledge representation and reasoning strategies that were designed to respond to the frame problem, e.g., frame semantics, even when they were not scalable due to too much hardcoding.
