Neil Brown6d ago

*Edit*: here at least, I am clearly not isolated!

Perhaps I am increasingly isolated in holding this position, but I have no interest in reading "AI"-generated slop.

I love reading.

I read people's blogs and toots and whatever *because people wrote them* and I want to read their own thoughts and opinions.

I buy books, and read numerous different authors. I like finding new authors, bringing new ideas, styles etc.

Same with "AI" images. I'd prefer no image at all.

Show threadLuc
@neil +1. The only place where AI generated text is a useful read is generated software documentation. It works pretty good for that.
Mar 15 at 7:02amWeb
Show threadKim Crawley 😷 (she/her)6d ago

@stroobl

No! 🖕🏻🖕🏻🖕🏻

Show threadLuc6d ago
@kimcrawley thanks for sharing you kind opinion.
Show threadNeil Brown6d ago
@stroobl I'd prefer documentation written by a human?
Show threadLuc6d ago
@neil the humans who made the software left and didn’t bother to write any. ;(
Show threadiwein6d ago

@stroobl
AI can write good docs is an unpopular opinion among people that know the value of documentation. Rightfully so because:

1. "AI" doesn't exist, and LLM's can't "know" meaning. They hide it well.
2. Writing documentation proves the author cares about the thing they document.
3. Documenting something surfaces it's subtlest bugs.

If devs use llm for documentation, the software is shit. If you use llm to describe others' software, you're taking a great risk of misunderstanding it.
@neil

Show threadLuc6d ago
@iwein @neil you should be aware of the risks is valid advice for any tool you use.
Show threadiwein6d ago

@stroobl I'm aware, but why is that relevant?

My conclusion can be rephrased in that frame as: The risk return ratio of using an LLM to generate documentation is always less than 1.

The rephrasing of the premises is left as an exercise to the reader.

@neil

Show threadmirabilos6d ago
@stroobl @neil fuck no