graydon-pub.bsky.social
extremely dubious of "everything is public" social network
375 posts
2,075 followers
302 following
Regular Contributor
Active Commenter
comment in response to
post
yeah or whatever the early program monitor / loader / OS was (often just a widely available fixed input tape / card etc)
comment in response to
post
i mean i think it kinda fits with the military origins -- people programming the things were just following orders themselves!
comment in response to
post
yeah totally. I use it selectively, as I can find things it's good at, and sometimes it blows up in my face. turning it off in many contexts is 100% the right choice. especially as a default.
comment in response to
post
(it is more or less based on a transplant of those same people's rhetorical treatment of blockchains, which they frequently view all "AI" investment as a simple pivot/redeployment of capital from. and there is an element of truth to that, the industry does love creating and chasing hype.)
comment in response to
post
yes, to paraphrase only slightly: a lot of FOSS people treat it as a scam that just generates pure garbage that people are simply deluded into thinking is real, while ruining the world and the internet and everything they hold dear, and therefore anyone who touches it is infected and dead to them.
comment in response to
post
eh, I mean, the FSF has spent years of lawyer time in GPL enforcement actions, it's congruent.
comment in response to
post
those are very reasonable objections! but they're far from the most acute I hear. a significant contingent are in full butlerian jihad mode.
comment in response to
post
to me the gap is like: suppose we all agree we want a substantially different economic system than the one we have. but then .. that's not forthcoming! what exactly do we do in the meantime? what do we try to organize for or against? it is perplexing. I have a wide tolerance for differing answers.
comment in response to
post
I think it is fair to ask people to look in a mirror if they've been espousing terrible politics forever but now the leopard is hungry for their face. but also it's not shameful to want to protect your livelihood and we shouldn't be too mean about that. everyone wants that, it's normal and good.
comment in response to
post
(part of me, darkly, thinks FOSS people are so critical of this because deep in our hearts we know we did a similar wage pressure commodification routine to proprietary people 20 years ago, and are either displacing guilt or can't stand the thought of being next. capitalism is terrible: news at 11.)
comment in response to
post
yeah, exactly. there is already a lot of code shipping at quality levels we don't care to discuss in polite company. not because the coders are necessarily bad, but the businesses are intensely cost sensitive. and if you replace those jobs with LLMs, you just move that wage pressure elsewhere.
comment in response to
post
Just as call centers with the common business model of "make the customer go away, at low cost" may be quite happy with chatbots RAGing up existing FAQs. Not a satisfying simulacrum of concierge-level customer service but the existing bar being cleared may be so low it's actually underground.
comment in response to
post
I especially think: for the part of the software industry that never ships a finished thing at all -- the "investor storytime" model of just getting rounds for a while then getting acqui-hired -- you might find even today's LLMs perfectly fit your needs for "crappy prototypes that change rapidly"?
comment in response to
post
like capital already does and will continue to use LLMs as a tool for labour discipline, to put pressure on wages, and to replace people at the margins. and since 90% of software never needed to be written anyways, maybe their delusional "business needs" will actually be met by that? who knows.
comment in response to
post
yeah sorry I was arguing with an imaginary critic not you. and I agree we're all still going to be programming in 12 months. or most of us. I think there'll be job losses but mostly ill-thought. the thing is people hire and fire programmers for lots of bad reasons. the labor market is very .. moody.
comment in response to
post
But I keep running into well read careful thoughtful people who absolutely hate the entire enterprise, have detailed essay-length debates ready about how it can't do what it does and never will because of fundamental philosophy reasons or whatever, and have never so much as tried one. It's so weird.
comment in response to
post
It's not "snake oil", it's the result of a half century research and massive industrial engineering project that incidentally kept the lights on in every CS lab you idolize for most of that time. And modulo quite a lot of caveats, it does stuff nobody has any other way of doing. It's not going away.
comment in response to
post
Do they fuck up a bunch? Absolutely. Do they do a ton of stuff that was 100% science fiction just a few years ago? Also yes. Are there big gaps still to figure out, and major socioeconomic and environmental costs to work through? Yes again. But come on, we got neural nets working! Like from the 60s!
comment in response to
post
I dunno some colleagues shared deepwiki.com with me and it seems like docs-from-code is a mode they do fairly well at.
I mean, after a while working with them I think a reasonable human can conclude they are forming _some_ interesting, nontrivial internal representation of their subjects. Just odd.
comment in response to
post
I see, I see (strokes imaginary communist beard pensively)
comment in response to
post
tbh I do not care as much if there are famous people on my social network; I kinda just want a cozy friends-of-friends network anyways. fedi is honestly working better for me. I think N:1 or 1:N communication fan-out or fan-in for very large values of N is typically a bad idea. especially un-gated.
comment in response to
post
tbh I do not care as much if there are famous people on my social network; I kinda just want a cozy friends-of-friends network anyways. fedi is honestly working better for me. I think N:1 or 1:N communication fan-out or fan-in for very large values of N is typically a bad idea. especially un-gated.
comment in response to
post
oh it is mean to leftists too! canonically, leftists never like to leave other leftists out of their critique. speaking as a leftist. the trick is to not set up a website everyone in the world gets to send you messages on. because many people are mean.
comment in response to
post
same
comment in response to
post
yeah nobody likes learning this one
comment in response to
post
sorry. good luck.
comment in response to
post
America is particularly bad about it. I think it's sort of in the origin story: a flight to escape aristocracy + a bunch of slavers / pirates / mercenaries + a bunch of religious cult leaders who thought weird god was talking to them.
comment in response to
post
gh merge queues are great. i would have killed for them back in the day. bors originally used buildbot for goodness sake. pull-in-a-loop model.
comment in response to
post
hey, the world is waiting for nextbench. could be your big chance!
comment in response to
post
if you look at other commits around that time you'll notice I was also trying to install something called the "metrics ratchet". This was a CI mechanism that would record and update performance metrics and refuse any regression. unfortunately, that never stuck.
comment in response to
post
turn it off. go back to push first and test at best-effort speed, accept frequent broken tree. that was the status quo for all software at the time. CI wasn't a gate. it was just a thing that tried to run tests as often as it could and send angry emails to everyone who committed recently on failure.
comment in response to
post
I was arguing to keep it, just parallelize the testsuite to speed it up. I wrote bors in the first place and it was still somewhat new and debated. I won that argument but lost the confidence of management.
comment in response to
post
Wonderful! I'm a huge fan of this book!
comment in response to
post
(hah yeah wow note the commit date -- I left the project the following week. that is literally my final commit to the project! and it seems like it only stuck for a couple years, was disabled in github.com/rust-lang/ru... and removed entirely in github.com/rust-lang/ru... .. oh well!)
comment in response to
post
yeah I put it into rust in 2013 (github.com/rust-lang/ru...) and uh .. actually .. fighting over this (and bors cycle times generally, and maybe turning bors off because it's too slow) was a thing that was one of a few last-straw arguments that ultimately ended my time on the project.
comment in response to
post
Huh! Now I am curious if this is convergent evolution or what -- I added sharded execution to LLVM's lit testsuite back in 2017 (github.com/llvm/llvm-pr...) and my hazy memory is that I was just copying something that we did in the rust testsuite since forever -- I thought nexttest also copied that!
comment in response to
post
fingers crossed too!
comment in response to
post
i just mean the egg and cheese schtick. maximum endearment.
comment in response to
post
I would vote the hell out of that ticket if I lived there