Summarization
Using a language model to condense a longer piece of text into a shorter one. It is one of the things current models do reasonably well, and it is genuinely useful for triage: turning a large, messy pile of recovered or archival material (for example, the output of scraping) into something a person can actually work through, or drafting first-pass metadata and finding aids for a collection.
The thing to keep in mind is that a summary is an interpretation, not a neutral shrinking. The model decides what to keep and what to drop, and it can drop the detail that actually mattered, smooth over a nuance, or state something the source did not. Like any language model it is probabilistic, so the same document can be summarized differently from one run to the next. For low-resource languages or culturally specific material it may also misjudge what is important, because it is working from patterns in other languages and other contexts. Summarization is best treated as a way to make a large amount of material navigable so a person can decide what to read closely, rather than as a replacement for reading it.