Genuine question, if you're a professional dev, how do llms affect your work? Do you utilize them often, or maybe not at all? If you use them, do you just occasionally ask it questions, or do you prompt everything and then just prove read the output?
>>107767684im a solodevi use llms all the timeespecially when im familiarizing myself with new stuffbut i also throw ideas at it and see what comes out the other endsometimes im lazy or im drunk and i let it figure stuff out for me, even if it boils down to looking up alternatives then comparing themthey honestly suck at programming proper, in anything that goes beyond a minimal working example
>>107767684i personally don't bother due to the confident-lying behavior and unpredictable lack of correctness in llm output, i find it tiresome and frustrating. but but maybe things have improved over the year and i'm missing out.some coworkers allegedly use the autocomplete or have used llms to generate unit tests successfully.if you make a simple repetitive change just right (within 1-2 commits) it may be able to replicate it in other folders/repos, but that means knowing the exact changes ahead of time.they bought the licenses a while ago, but no one is really that wiser how the codebase works or more willing to do the dirty work to make it better.there is certain aura of desperation in engineering mgmt trying to find and demonstrate a productive use of llms.they just outsource more people when they notice they're falling behind in delivery. training those people takes time of course.
>>107767684I am Kip now from Napoleon Dynamite
>>107767684If nothing else, they're pretty good at generating tests that you can't be fucked to do yourself. Also, provided you have well-documented tasks you can throw them directly into something like copilot and get decent scaffolding for a solution. It probably won't work right off the bat and will miss several edge cases, but it's a good base to start working off of.
>>107767783>confident-lying behavior and unpredictable lack of correctness in llm outputthis is the worst part about it. it will just lie to your face or over exaggerate to a degree where it becomes painful to experience it. if i need to use it i have to give it so much detail so it doesn't go off rambling about things i didn't ask for.
Senior LLM Dev Only time I ever touch the companies language model is to fix the errors that the code monkeys can't fix
I’ve been using it to rubber duck debug shit I coded manually and give me suggestions. As in AI as a partner I’d have when doing XP or something because people seem to think pair-programming is bad nowadays.
I switched to dedicated LLM coding editors (Cline and Roo Code VS Code) and it's much better than just copy-pasting from the web interface. It genuinely accelerates my work now. Gemini 3 is a beast.I don't get people who say LLMs are not useful because they sometimes make mistakes or hallucinate. Reading the produced code, using a linter and compiler takes care of 95% of such mistakes. Running tests and occasional debugging takes care of 99%. I cannot remember the last time I got a non-trivial bug due to LLM hallucination.
>>107769017What about the convoluted logic and syntax what often keep changing even when it already has lots of context? Why would you care if it is functional or not if it is pain to decipher anything what the machine has created..
>>107769017For jeet spaghetti vibeslop this is more than sufficient.The one thing LLMs still and forever will spectacularly fail to do is reduce the amount of code.
>>107767684I'm too lazy to bother to integrate it with my workflow.
>>107767684>f you're a professional devUsed to be.>how do llms affect your workTook my jerb.>Do you utilize them often, or maybe not at all?Never.I felt that that I have a responsibility as a developer to produce the best code that I can. Which means I need to understand everything I do, which means I can't have some slop generated for me. But that made me very slow. And now I'm unemployed.Software development isn't for me anymore. Maybe it never was. But t was the only thing I've even been remotely good at. Apparently I'm not good for anything anymore.
>>107771257Holy shit are you me?
>>107771257You should have used LLMs if you didn’t want to be left behind. Remember, they don’t care if you enjoy programming, they want you to produce stuff for them, the faster the better
>>107771472I think you only read half my post. This reply adds nothing new to what I already said.
>>107771575You are trans and Luddite then
>>107767684>be me>be professional of 20 years>can code blindfolded>write a function>give it to the LLM and ask "who wrote this code?">"Sir, clearly a white person wrote this magnificent code. I cannot find any faults in it, unlike the shit brownoids give me">"Thank you, Claude"
>>107767684They help me and others write some deocumentationslop nobody will read in confluence.Also very useful for coding, snippet suggestions can be hit or miss, but most likely a hit with minor adjustments.It's also useful in order to brainstorm some ideas and see what people have already done similarly in similar scenarios.
>>107767684I don't use them. I know some of my colleagues do, to various degrees. I still haven't been fired for being less "productive" than them, so I think I'm good.
>>107767684I replaced Google with ChatGPT almost completely.
>>107771257Using LLMs doesn't reduce the quality of your code unless you're too lazy to review it.And LLMs aren't capable of taking dev jobs yet. Maybe you just sucked.
>>107767684I have to shoehorn them into what I am doing. It reduces productivity a lot, but middle managers who've drunk the koolaid are tracking AI usage now and tying it to performance reviews. If it really made engineers more productive, then it wouldn't need to be measured explicitly. The more productive engineers would accomplish more and and it'd be obvious.
