I find using LLMs for summaries of news headlines to be quite helpful. Many headlines in my RSS feeds have turned awfully "clickbaity" over the years. Using an LLM at the individual article level, it's usually a pretty great way to produce a new informative headline from the full article body. It makes the information much more relevant, and better allows me to decide whether I want to read the full article or not.
(Just as an example of practically useful functionality, while I agree that the technology still falls short of expectations and dreams in many other domains)