Gregory's Server
 100 comments
 One bit of advice i was given as a teenager: "Always have a fallback plan that's profitable, but that you really hate." "That way, you'll do what you can to avoid using it, but it's there if you need it." "When you look at your current situation, if it starts to look worse than your fallback plan, it's time to be somewhere else..." :D   @middleclasstool @phire took me a minute to interpret bussing as a verb and not an adjective like I realize I intentionally self select for programming friends who are commie pinkos who hate everything the current hype cycle represents but there aren’t so few of us even in the centre of the tormentus nexus enclave that we’d constitute anything smaller than a significant minority, and I am guessing the prevalence of things like copilot among non-SV devs is lower, not higher, just based on typical tech adoption curves if nothing else  @phire I've observed two broad moieties of LLM based code tool user: First the extremely skilled and knowledgeable, who do not benefit from them, but whose expertise allows them to piss away enormous amounts of time playing around to see what they can be made to do. Second, the very low-skilled practitioners who lack the ability to assess the results, and therefore wind up making a mess they cannot dig themselves out of. Both groups waste time, effort, and carbon emissions for no net benefit.  I'm sorta in the first camp, but I'm not particularly skilled or knowledgeable but I've worked with neural nets since the first practical vision systems came along. The problem resolves to what an LLM can do for anyone. That's where the structured programmer stumbles badly. Most of these guys have never used the volatile keyword in a declaration. What they have to say about a technology they don't understand is a waste of calories. @creachadair @phire I’m mostly in the second camp - but just good enough to modify, debug and integrate the outputs. Thing is; I’m not full time dev, or even 25%. Sometimes I go months without coding and then boom I have to fix something or have to satisfy a new demand. I’m NOT a real dev, but there are probably more like me out there. @phire I think the most reliable number I can find is from Stack Overflow’s 2024 survey, in which about 62% of respondents indicated they were using AI coding tools. https://survey.stackoverflow.co/2024/ai (SO has further details on their survey demographics which I would personally assume bias this number in the upward direction, although I don’t know for sure.)   @phire many software dev companies have specifically banned copilot and similar for any dev task that deal with changing older existing code. In our experience it tends to slow down devs because they spend more time finding and fixing bugs afterwards, but fully understand less of the code changes, thus taking longer to understandin depth what is happening.  I, professional programmer, have *zero* peers that use AI "assistants" professionally. (That, or they're hiding it from me.) @suetanvil @phire same. this feels like AI fanboy hype bullshit, I don't know anyone at all competent at their job who thinks AI "assistance" is a good idea.  @phire I know of basically no devs who uses AI coding assistants, at least not on purpose and in a sustained manner. Once in a while I’ll read a True Believer™ claim amazingness on the company messaging tool, but really, everyone is collectively shrugging their shoulder and going back to the usual coding tools.   @drahardja a couple times I’ve tab completed my way through the boiler plate of a unit test, but that’s about it. @waffles I love it when I type “enum” and my IDE offers up a block of constants from some random project. Sure, buddy, that’s exactly what I want. @drahardja oh man, ok i wasn't even considering copilot trained on random public repos. i've only ever used the internal version that my work uses, that's trained on our mono repo -- which basically accomplishes what i would've done otherwise: copied my deskmates work :P But joking aside, i don't personally find those tools helpful. I'm happy for others who do, but it's just not my thing.  @phire I’m as far as you can get from commie pinko, but — so far — even on purely utilitarian grounds, I haven’t found these things to be of enough benefit to use with any regularity. They’re far too mistake-prone.  @phire Here's how it looks from my torment-nexus-adjacent workplace: Junior devs: mostly ignorant of the threat, mostly some adoption.    @phire Let me see now ... Somebody who uses such a thing will no doubt fail to understand the "code" it has "generated" for them, on account of not having understood the problem in the first place ... ... so won't be able to answer code review questions from a reviewer who doesn't understand the "generated" "code", on account of it being garbage ... ... ah, but, oh yes, of course they will ... ... all they have to do is put the code review question into ChatGPT and pass on the answer, sorted.  @phire I've made extensive use of Copilot, Gemini, etc. to ask them simple questions and be given extremely verbose obvious lies. Does that count?  @phire yeah, AI is an ongoing joke at my workplace. The only people using it regularly are writing boilerplate docs; the devs post anti-AI memes and favorite bonkers AI output in the #shitposting channel every day.   @phire Copilot-type? No. Xcode offers it as an optional package but I never activated it. I would certainly avoid activating AI completion features at work because if it ever became a thing we used for whatever reason, it would need the employer’s approval first. I’m not confident that auto-completed AI code is something I would enjoy. I think coding in itself is rather fun: why would I ruin that experience? Also, I need to understand every code line before calling it ”my” code.  @Arcticulate @phire I avoid AI assistance like the plague. A terminal I've used for years on macOS iTerm integrated it, and I was just horrified that anyone thought that's a good idea. It's bad enough we `curl $url | bash` but to elevate AI's answers to one key from execution—NO, THANK YOU. So I shopped for new terminal program and found wezterm, and it has cat-able images! https://mastodon.gamedev.place/@shanecelis/113259541823069873  @phire They would be so fired. To get the LLM to do anything useful, they’d have to feed the company’s proprietary, copyrighted code base into a server-hosted LLM that doesn’t know how to forget what it just read. The risk that code “escapes” is the reason their employer is not taking the chance. Doing so without permission would really bother the lawyers. @phire That assertion seems highly unlikely to me. We use visual studio but no one on my team uses copilot. There is a new management push in our office to use it but I am waiting to hear their justification. I remain very skeptical. @phire I'm gonna go out on a limb and guess that their profile says 'Elon is the Einstein of our generation"  @phire Yeah that's complete horseshit. My employer would be catastrophically disappointed in me if I would let outside code leak into our proprietary code base. And on a personal level, I love programming for the whole writing code part. People who hate coding so much they need a machine doing it for them might not be in the right career as programmers.  @tsturm @phire There's also an argument that because LLMs are regurgitating high-probability token streams from large datasets, you are getting at best average code for simple, common tasks. Which means if you find copilot-like tools useful, you are possibly a below average programmer doing scut work. @overeducatedredneck @phire And the code might actually be better than whatever I've cooked up and re-used over the years, but I still prefer re-solving some JSON parsing problem for the 10th time, because maybe this time - finally! - I understand where I went wrong the last nine times. 😅  @phire we tested it internally and AI-support is dogshit. Many programmers are not ready to use anything newer than IRC and you want to tell me "most" programmers use AI already? From the deepest of my heart let me help you: public coping is not effective marketing. Publicly lying to yourself hoping to "make it til you fake it" is extremely transparent and cope just doesn't look that good on you.  @phire Might be technically true. VisualStudio, for example, has AI integration - doesn't necessarily mean that anybody uses it. Same with Windows 11 etc  @phire omg con{grats,dolences} on the Gargron boost 😭 I could see "majority of programmers have tried AI at least once." There's a lot of hype, it's widespread in some circles. But "almost every programmer" using it regularly? No way. Some of us are staying the hell away from that shit, even have a hard line against working for AI companies.  @phire 92%, per this survey: https://github.blog/news-insights/research/survey-reveals-ais-impact-on-the-developer-experience/ I don't personally know ANY software engineers who refuse to use AI. Some at work disliked the inaccuracy of Copilot and disabled it, but they're not ruling out AI as a whole.  @randytayler @phire Well, we (programmers who refuse to use AI for everyday coding) are out there, you can find a bunch of us in the replies to the OP here. I'll check out the link later but a priori I'm skeptical that that 92% figure is accurate. The 62% someone posted elsewhere in the thread sounds more plausible.   @phire I used to use it inline all the time, I don't use it as much anymore, only in the chat for refactoring and converting openapi to Go. @phire Hello my company needs a tutor who can translate English to native native language just 50pdf words Send a message on telegram via this link! @phire Funny stuff. That's some real echo chamber BS. Absolutely none of my team uses AI. It generates garbage code. The loss of quality would be immediately obvious in code review. The perpetrator would be mocked and probably fired. Come on, kiddos. Just do the work and stop cheating. These are the wrong use cases for AI anyway. The right use cases are idea generation to start a project, not code to implement a project. @phire@phire.place where can I find an employer that doesn't encourage it? That sounds like an awesome employer.  @phire Out here doing my part as one of the few developers who is more efficient when I’m not code reviewing AI shite.  @phire the only way this could be true is if they're counting how all our IDEs now include this bullshit by default in a way that can't be opted out of, and also if by "using" it they mean "swearing at" it for constantly inserting itself where it's not wanted (everywhere) @phire What does that even mean? Maybe student programmers with zero experience are using copilot-type products? Cause I would be happy to show them, with 35+ years of commercial software development under my belt, not only am I not using any sort of AI to do my work, but I strongly doubt that today's AI could do my work. (I build neural networks in C++ to do object detection.)  @phire I used it once, it saved 20 minutes at first, wasted 15 minutes afterwards, and I never touched it again. Granted, I wasn’t asking it to write me an algorithm, but rather gave a full verbal description of a class signature and test cases I was interested in seeing. After third clarification prompt it started losing context, and I ended up rewriting most of the scaffolding anyways. And well, being a TDD freak, I simply don’t see a use for copilot-like tools in anything more complex. @phire Unfortunately since all of the forums are polluted with stealth AI garbage this statement might be more true than I’d like to think.    @phire I refuse to because I actually like programming and I explicitly do not like fixing other people's shitty code, so why tf would I want to fix some code conjured up from the ether by a black box plagiarism algorithm? @phire well. I’m in! Also, my 17-year-older than me lead used AU too. I’m 75 and he’s 90 and you want more data? 🍸😿🍸🙀Gimme an effin break!🍸😹 BTW it’s not super great duh! But you had to know that right?🍸😹  @phire have not, will not. Disabling “AI” wherever it intrudes into my work and life. It’s bullshit. Every single coder I've ever met just felt a great disturbance in the force, as if a terrible lie was told about AI yet again... |
@phire I'd literally rather go back to bussing tables