Regarding LLMs and summarisation: many of the more important summarisation benchmarks that are used to guide model training, as well as the most common use cases, are built around papers and news articles, both formats that already contain their own summaries (abstract and first paragraph, respectively)
This helps explain why attempting to use these systems to summarise other forms of writing like legislation, regulation, or books, they often fail miserably with extremely incorrect statements.
Because people who like stuffing their brains with all sorts of top level familiarity are convinced that they're getting deeper understanding from summaries, after some point of confusion through they themselves summarizing primary sources in their own mind. The cognitive process of self summarizing is not the same as reading someone else's summary to gain understanding, but they think it totally is after doing so much self summarizing and knowing its a slog.
Keen example - motherfucking dorks at UW in the brain science department creating podcast summaries of papers with AI so they can get a passive understanding payload on their way to work. My Dad was touting this as the apex of the current utility and I looked him right in the eye and asked if nerds love tools that reinforce their nerdiness.
Baldur Bjarnason
Leo Ré Jorge
antimagic heart-sickness
Ciggy Bringer of Smoke