Post

apparently hot take: it is far more important for the internet archive, government services, and other critical pieces of public infrastructure & the commons to continue to function than people to feel 100% morally pure in their consumption. there's certainly a space for moral stances, but...

... "backporting a critical patch to an unmaintained-but-core-piece-of-infrastructure before serious damage is done" is not one of them.

am i allowed to take the moral stance that "we fed an 11k-line patch to the magic automatic intern and then asked it to gauge whether it did a good job" is not a reassuring response to a security issue

that is ur perogative, but that's definitely a different stance that im talking about here. I expect ur also significantly more informed about the difficulties patching older software than most folks in the original thread.

(that does seem more like practical concerns than moral ones, tho 🤷‍♀️)

hm, depends whether security devs have any moral obligation and i'd guess the sentiments all trace back to some flavor of "this thing sucks up people's hard work to produce endless junk that we're constantly assured is just as good". not many people object to the water usage of a thing they love

> hm, depends whether... fair, i think it's reasonable to argue security devs have a moral obligation to good security practices (that was not originally how i interpreted ur words), but i also suspect ia gave it more thought than just "yolo a patch" …

> and id guess... im not convinced this is actually the case given that a lotta folks' sentiment doesn't seem to vary based on the perceptible quality of the output, or when the output isn't perceptible at all, which is a core part of my frustration here; there's a vast difference (to me, at least)…

…between "digital littering/pollution is gross" & "anything an llm ever touches, no matter the amount of care or the practical outcome, must be rejected upon first discovery of its taint without further consideration", the latter of which seems to be fairly prevalent & which which i take issue.

again, i do actually think much of what llms produce is trash! including some of their use in software eng! but i do think there's cases where their output is not measurably worse than humans, & in certain cases they can enable humans to move significantly faster, be more robust, or locate…

…obscure information more easily, and i think it's important to actually evaluate those cases as they come, on their actual merits & flaws (which can include side effects like energy usage, etc)