dynamic

In whatever cases LLMs are being used to generate so-called "boilerplate" text that actually servers no function, the correct solution is *not* to outsource human thought to a statistical model, but to reduce the volume of text generated overall. 2/3

But what if, just maybe, some of the things we think of as "boilerplate text" that "no one reads" are actually occasionally extremely important. Think about incident reports following accidents in factories, documentation of medical practices, or judicial opinions (see: https://washingtonpost.com/nation/2026/04/02/judges-ai-hearings-rulings/) Sometimes the boring stuff is important. 3/3

@MastodonEngineering@mastodon.social It would be helpful if these questions were less mandatory. I would *really* like to skip question 2 because I do everything I can do to avoid being influenced by algorithmic recommendations.

@agoldst@mastodon.social Thanks. I also like the immediately previous quote: "...we know (quite spectacularly, from the world of self-driving cars), that it is very very difficult to keep paying attention when you’re monitoring a system that is meant to be A-OK most of the time, but needs you to be constantly aware because it sometimes screws up in a way that requires immediate action." #LLMs #genAI

@Nathanpboston Wish I had something more constructive to say than "ugh." : (

@paulcarbuncle @jesse Thanks. This person offers the same explanation, and includes a citation: https://cohost.org/censa/post/296287-empty

@benjojo@benjojo.co.uk Interesting questions. Any way to get added to a list to be notified of your results?

@elCelio@mastodon.uno @MastodonEngineering@mastodon.social That's not what the poll was advertised as doing.