The first thing to do is to put LLMs to work to generate large knowledge bases of commonsense knowledge in symbolic machine-readable formats that Cyc-like projects can consume.