Interesting quote towards the end... > [Doug] said, though, that he found it “increasingly worrisome that these models [LLMs] train on CONVINCINGNESS rather th

Interesting quote towards the end... > [Doug] said, though, that he found it “increasingly worrisome that these models [LLMs] train on CONVINCINGNESS rather than CORRECTNESS” That's quite a neat encapsulation of the difference between CYC (or what I understand of it from that article) and today's GPTs.