A very thorough and interesting article -- probably the only thing you ever need read about Cyc. An example of where it shines is the way it carefully pinpoints Lenat contradicting himself on the question of converting English into Cyc's internal form -- Lenat claims that this could likely be done with NNs, but also that NNs are not capable of "common sense", which he himself (correctly) claims is necessary for language translation.

It's not without some flaws, though. The continual harping on how secretive Cyc was doesn't seem fair, given that (with the improbable exception of Meta) none of the currently leading NN-based LLMs have revealed their "source code" (weights) either. Also, I don't agree that Lenat's single-mindedness was a bad thing (I'm glad somebody explored this unpopular concept), nor even that he was necessarily wrong in believing that building a huge, curated fact base is a path to AGI -- the graph showing the exponential increase in the number of assertions in Cyc over time, despite roughly constant headcount, points towards the possibility that a critical threshold was nearby. In fact, the history of NNs is eerily similar: at first found to be promising, then abandoned, then -- against all prevailing wisdom at the time -- found to actually perform very well if you simply scaled them far more than you "should".