[CCing groff list as it's also on-topic there, as troff history is to
both lists]
At 2025-05-25T10:59:13-0700, Adam Thornton wrote:
> I don't think I'd place that kind of faith in the LLM summary without
> having a good read first.
[...]
At 2025-05-25T16:13:58-0400, Norman Wilson wrote:
[...]
> LLMs are not search engines. They are bullshit generators.
I had a similarly bad experience with troff history and the AI-generated
"answer" Google situated in prime visual real estate.
This "AI Overview" claimed:
"In troff, Brian Kernighan modified the text formatter to include a
sentence space, which adds extra space between sentences. Specifically,
if the sentence space is not set to zero, the second space after a
potential end-of-sentence character (like ".", "?", or "!") will be
treated as a sentence space, adding extra vertical spacing."
My aggrieved rant, which some of my social media friends endured:
"Jeeeesus Christ. I needed to look up some advice Kernighan wrote to
troff users decades ago and I got a gout of lies sprayed directly into
my face in an area where I might be able to call myself a domain expert.
Be VERY VERY careful accepting any AI-generated results, people. This
shit is ALMOST COMPLETELY WRONG. (It's correct about which punctuation
marks [can] end a sentence...but you already knew that part, didn't
you?)
1. Kernighan didn't innovate this. Ossanna troff (and nroff) already
implemented end-of-sentence inference.
2. Configuration of inter-sentence spacing is a GNU troff extension,
something else Kernighan had nothing to do with. Ossanna and Kernighan
troffs permitted configuration of the inter-WORD space size with the
`ss` request, but not the inter-SENTENCE space size.
3. Inter-sentence space is NOT VERTICAL. It's horizontal."
Some things it's quite good on. Other things it's terrible but plausible sounding like this. You never know which you'll get and as a non-expert, you can't know which is which.
Warner