x
Verified Knowledge Extraction from Language Models: relations as operators, a graph as the gate, and three retracted results — LessWrong