@techokami > The end result is an order-of-magnitude speedup for a read()-heavy benchmark. Review has already identified a race condition in the code, though, so further work will be needed.
every time
@techokami > The end result is an order-of-magnitude speedup for a read()-heavy benchmark. Review has already identified a race condition in the code, though, so further work will be needed.
every time
@linear @techokami I love how hard the LWN comments are trying to talk over how damning this result is, including accusing the author of uhhh not glazing torvalds enough?
@techokami @linear god, he really is transforming before our eyes into the LLM guys at work who think their job is to shift blame to their coworkers and always find a way to make their mistakes someone else's problem in review
@zzt @linear should post the exact message he did this in: https://lwn.net/ml/all/CAHk-=wiCe2H33izZHDyDHEhcFwUr9fXZk27-WZ2vn_pn5FcZTg@mail.gmail.com/
@techokami @linear I can't emphasize enough how much all of this, down to pretending prompting is a skill and it's everyone else's fault for not picking it up, and being overly impressed with basic editor tasks because the LLM did them, is all shit I've seen from LLM addicts in a commercial setting
Just wait 'til they rediscover regular expressions.
@zzt @techokami @linear he always have been. This is only a revelator bath.
@techokami @linear @zzt RCU is a huge mistake, one of the first changes that made the kernel vastly worse than it used to be. Nobody understands it or has a good formal model for correctness of it, and it pays in late-failing resource allocations (= no way to recover under resource exhaustion, boom!) as a tradeoff to eliminate logically-necessary locking, for the sake of performance on highly-parallel loads on highly-parallel machines. For us normal folks, it's just worse in all ways.
@dalias @techokami @linear @zzt ah I though I found this shit overly complex because I'm just a systems noob.
@dalias @techokami @linear @zzt The RCU sounds like the kind of thing that shouldn't be written without starting with formal methods and a full TLA+ model.
@techokami would you be surprised to know that this is more or less how all the other vibecoded systems projects went too? there’s a scathing code review of anthropic’s C compiler out there
@zzt
He must've been prompting it wrong.
@dzwiedziu @techokami bbbut his bulletproof prompt that only a skilled programmer could’ve come up with (no you can’t see it)
@zzt @dzwiedziu @techokami This is why I only program by using an OUIJA board to pick page numbers from the C reference book.
@discatte @zzt @dzwiedziu quote of the day
@discatte @zzt @dzwiedziu @techokami I only program using an OUYA game console. No you can't know how I get the controller to connect consistently, serious developers only. If you're poor every compiler has a free demo.
@zzt @techokami lol, lmao
@zzt Guess i'm taking a second whack at sorting out the X driver problem on my FreeBSD install ...
the classic 'we can make it WAY faster if we don't need it to work, but trust me I'm 90% of the way there now'
@Patashu @zzt @techokami we all know the last 10% take 90% of the time, but with llms, it's much worse.
@zzt @techokami "thats fine, the holy orange asshole will fix it when asked politely"