RE: https://mastodon.sdf.org/@argv_minus_one/117207930196699334
yeah, so when LibreOffice repudiated the AI, that was what we call a lie
0 Boosts
0 Likes
@davidgerard@circumstances.run @atax1a@infosec.exchange
The answer to how much code the bobot should be trusted to write is zero. In Dutch, nul.
@davidgerard JFC. They used an LLM to write 24 lines of code AND STILL HAD TO DEBUG THEM.
“The copyButtonHandler method was manually created and had
support from LLM/code completion. (lines 330-353)
The patch is open to further suggestions for simplification
and code improvement or refactoring.
Some lessons learned: LLM cannot guess solutions for wide
and open demands. Narrowing the query to code fragments
produces acceptable and easily debuggable code.”
@LabSpokane @davidgerard the "using slop for debugging" demo at our day job had a guy sitting there telling us "uhhh, the model you choose sorta doesn't matter, and also you have to tell it where the bug is"
totally useful technology, love having to pay money to sam altman to do debugging by telling a chatbot where the bug is, makes lots of economic and logical sense
borat voice not
@LabSpokane it really shows that Collabora was doing the work in the project
@LabSpokane really hope they haven't gone slop
@davidgerard The comments alone should make the case, but we’re in the stupidest of all the stupid times ...
@davidgerard What does the post say?
@kimcrawley that's three openly chatbot patches
@davidgerard@circumstances.run
Their response to being called out for using the bobot (I'm blaming @atax1a@infosec.exchange for me learning of that word as they're the first person I remember using it) is exactly as bad as you'd think.
https://fosstodon.org/@libreoffice/117214068793979843