You could compare it to T5 instead. T5 does awesome at trivia, both the code and trained models are available, and it'll run on ordinary hardware. (Not sure if

You could compare it to T5 instead. T5 does awesome at trivia, both the code and trained models are available, and it'll run on ordinary hardware. (Not sure if one could argue 'the millions of dollars' part, but that GB R&D wasn't spent for the purpose of trivia playing like the Watson Jeopardy system was aimed at doing, so seems reasonably fair to me - the trivia playing is merely the byproduct of the target capabilities of translation, knowledge graph construction etc).