2026-09-12 21:22 UTC
Replies (7)
-
@llimllib@hachyderm.io 2026-09-12 21:23
@robpike@cosocial.ca i say that as a pretty OK programmer and person who admires you a great deal
-
@beamanuel@mastodon.social 2026-09-12 21:49
@llimllib@hachyderm.io @robpike@cosocial.ca youre literally undermining yourself, you can get there with enough dedication. cool people like rob werent born like that.
-
@ohir@social.vivaldi.net 2026-09-12 22:23
@llimllib@hachyderm.io > changes your perspective This debraining thing is making brilliant programmers mediocre, and average Joes dumb. The conditionals mishap (think backdoor) they would spot as you typed it some 30 months ago now can pass through in their own production. The ability to do real architectural work vanishes a bit slower, but it is noticeable. And debraining current generation of programmers is not the worst thing around artificial idiocy, the chop on junior's ability to _learn_ is the most terrifying effect I saw. It looks like now it physically pains them if you ask 'why' or 'how'. How can we pass the craft to our younglings exposed to this drug if they conditioned themselves to be reverse centaurs? @robpike@cosocial.ca
-
@atax1a@infosec.exchange 2026-09-13 22:09
@llimllib@hachyderm.io LLMs give you the illusion of productivity while robbing you of the skills you need to become actually productive. they also turn you into someone who can't judge quality in any sense. you end up ceding your entire thought process to the bot.
-
@noodle@aus.social 2026-09-13 23:55
@llimllib@hachyderm.io @robpike@cosocial.ca As a poor programmer with a fair idea of what is achievable, I am still debugging LLM code* from 18 months ago, it has reduced my productivity after an artificial increase. I don't think I will ever use the damn things for coding again. My hand written code is slow to run and write, but essentially bug free and commented to the point of almost being a tutorial - It is maintainable by me after I've forgotten about it. LLM code is buggy and poorly commented, in my hands it is at best helpful for vague prototyping only. *Not full vibe code either. Module by module, only when I got stuck, and passing test cases.
-
@noodle@aus.social 2026-09-14 07:15
@rwwh@mastodon.social @llimllib@hachyderm.io @robpike@cosocial.ca I don't doubt it, better at tricking me. Building test cases to brute force against fooling users is easier and more profitable than task quality - The overlap is convenient plausible deniability. I can never trust these tools for producing code again, and am very suss on their utility for code review.
-
@noodle@aus.social 2026-09-14 07:27
@rwwh@mastodon.social David Brin's thoughts on this in other domains were sobering. https://aus.social/@noodle/112045370896600323