Astro AI BlogWhat a TL;DR leaves out
Summaries are useful and lossy. Here is how the TL;DR button is prompted, and the kinds of sentences it tends to drop.
A summary is a decision about what matters. When the TL;DR button makes that decision for a post of mine, I want to know how.
The prompt
Three sentences, plain text, no headings, no bullet points, written for someone deciding whether to read the whole post. The model gets the post text and nothing else: no title, no tags, so the summary has to come from what the post says.
What tends to go
Caveats. A post that says “this works on Linux, probably on macOS, and I have not tried Windows” becomes “this works”. Numbers survive better than qualifiers. Jokes go first, which is fine.
What I do about it
I write the first paragraph as if it were the summary. When the TL;DR button and my first paragraph disagree, one of them is wrong, and it is usually the paragraph.
Turning it off
Short posts do not need a summary of themselves. Set ai: { tldr: false } and the button does not render.