I've found that the larger open-weight AI models do a great job of explaining the old non-fiction content on PG, particularly magazine articles which are a good size for the AI to handle. It breaks down the long wall-of-text paragraphs for you and explains all the historically relevant background that would've been assumed to be known back in the day.

If you ask it to assess the relevance of the text in the present day it will also do that very nicely, highlighting the places where the text shows old-fashioned viewpoints that would be sharply criticized today.

so maybe Karpathy has a point that LLM-assisted reading should be a thing. Would be cool if that worked on E-Reader screens as well. Maybe when the browsers on E-Readers become good enough ...