nevetIdentifying the Actors: What a Byline Certifies | Hacker News
dreamreal(I think the article's written well, but it ain't short.)
GreenResponsepersonally I kill AI with fire wherever I find it. Good thing Firefox has a killswitch for that
GreenResponseI hate this AI hype
dreamrealThat's dumb and impossible
dreamrealbut I get it
GreenResponsebut I realise that it’s a godsend for technical people that have problems with articulating their thoughts, pretty much most of the people
dreamrealyou're in good company: you're doing the same thing headius did
Paradreamreal: fwiw I recommend using Co-Authored-By trailers in commits with some pseudoidentity for LLMs et al. This pairs well with "commit in small pieces, commit often". Of course it's still a human process and most don't even really know how to add commit message trailers but I don't see any better way of doing it at this point beyond some completely hands-off commit authoring process with a
Paraseparate diff-by-diff signing process.
dreamreali don't see the point
dreamrealI mean, I do, but really... why
ParaRequirements are in flux, my previous project where my involvement just ended had a LLM authoring policy change where they do want people to clearly highlight which content came from where and I'm happy I don't have to solve that there :)
ParaThe trailers is a nifty and underutilized feature IMHO in general, although they're more useful for the reviewed-by and other such metadata which they usually are used to.
dreamrealI guess I understand the need for provenance, but even so: what does it DO?
ParaIt's a good lightweight solution to red tape.
dreamrealif the code works - like, tested, the verification is present - then why does it matter?
dreamreal"oh no, that line came from qwen, THAT one from gemma4, THAT one from... oh, it's from Sam, ew."
ParaFingerpointing/blame, I'd imagine.
ParaHow dare you make mistakes etc.
* dreamreal is so overjoyed at being reminded that people are stupid
dreamrealI can't wait until the world realizes that I KNOW THIS ALREADY, being stupid myself, and don't need further reminding
ParaOr Totally Accurate Stats such as "70% of bugs involved ChatGPT"
ParaOr Even More Relevant Stats such as "55% of production critical bugs were reviewed by John Random, fire him!"
ParaTBH while I'm the kind of metadata fan who likes this stuff but I can see lots of abuse potential for all these, but alas...
dreamreali don't mind the metadata, I guess, I just don't see how it's actually useful
dreamrealit can be interesting
dreamrealbut to me the metric is "does the code effin' WORK" and "do we know why we can make that assertion"
ParaOn that I'd rather use the metadata for adding e.g. CI process marking commits as Tested-By or somesuch.
ParaLike, since it's possible, why the hell not.
ParaOf course this means hand-over and history rewriting which will make the human part harder.
dreamrealWell, tested-by still relies on HITL, because the specs don't necessarily factor in properly
ParaIt's better suited for manual testing, sure.
ParaAnd in a sense that makes sense even. But then one needs to teach the manual testers how to do it or build a "click here to mark this tested" minitool or whatever...
ParaAll doable, and kinda very interesting, but still...yeah I don't know what the point would be :)
dreamrealEw, not THAT kind of manual testing
dreamrealI mean, that's good, QA people are worth their weight in gold, but ... ew