(评论)
(comments)

原始链接: https://news.ycombinator.com/item?id=43446695

这篇 Hacker News 帖子讨论了一篇题为“初级开发者的复仇”的文章,以及 AI 编码工具的使用。一位用户建议投资 API 积分并使用像 aider 这样的工具来评估这项技术对编程的潜在影响。其他人则争论 AI 辅助编码的实用性和成本效益。 一些人认为,代码补全工具对于特定任务仍然很有价值,尤其是在需要详细控制的情况下,将其比作“机枪”与“导弹”。他们强调,只有当用户拥有扎实的知识并能够引导工具时,AI 才能更有效地工作,就像操作重型机械一样。一些人还发现大型语言模型 (LLM) 不好用,因为需要不断地进行规范和验证。 怀疑论者质疑 AI 是否能够取代真正的编程,强调在编码之前思考和规划的重要性。他们发现大型语言模型效率低下,因为与直接编写代码相比,它们需要付出更多努力来规范、验证和编辑。一条评论将术语“vibecoding”归于 Karpathy,而另一条评论建议阅读 Bengio 等人的著作以进一步了解 AI。

相关文章
  • 初级开发的复仇 2025-03-22
  • (评论) 2024-06-13
  • (评论) 2025-03-11
  • (评论) 2025-03-14
  • (评论) 2025-03-22

  • 原文
    Hacker News new | past | comments | ask | show | jobs | submit login
    Revenge of the Junior Developer (sourcegraph.com)
    12 points by ado__dev 2 hours ago | hide | past | favorite | 7 comments










    I’m usually one of the people complaining about hype cycles, and it’s usually been correct to be pessimistic about them.

    But in this particular case I have to think a lot of people just haven’t tried it in its best form. No, not a local model on your MacBook. No, not the web interface on the free plan. Go lay down $300 into API credits, spend a weekend (or maybe two) fully setting up aider, really give it a shot. It’s ultimately a pretty small amount of money when it comes to figuring out whether the people who are telling you there’s an existential risk to your livelihood on the horizon are onto something, don’t you think?



    $300? That much would last a year. You don't need anywhere near it just to try things out


    I love how new technology becomes like religion. It develops both cult followers and critics.

    In that lens think the AI cult is more right than the Crypto cult. At least I can use it to do something tangible right now while crypto is still pretty useless after many years.

    In some sense I think these technologies need the cults and the critics though. It’s good to have people push new things forwards even if everyone isn’t along for the ride. It’s also good to have a counter side poke holes. I think the world is better with both optimists charting new paths forwards and pessimists making sure they don’t walk right off a cliff.



    > I have bad news: Code completions were very popular a year ago, a time that now feels like a distant prequel. But they are now the AI equivalent of “dead man walking.”

    I disagree. I view this as a "machine guns versus heat sinking missiles from the 70s" dichotomy. Sure, using missiles is faster. However, sometimes you're too close for missiles. Also, machine gun rounds are way cheaper than missiles. However, when they first came out, missiles were viewed as the future. For a while, fighter jets were made without machine guns, but they added them back later because they decided they needed both.

    Sometimes I find I want to drill down and edit what Claude generated. In that case, copilot is still really nice.

    With regard to ai assisted coding: the more you know what you're doing, the more you know the code base, the better result you'll get. To me it feels like a rototiller or some other power tool. It plows soil way faster than you can and is self propelled, but it isn't self directed. Using it still requires planning and it's expensive to run. While using the tool, you must micromanage its direction, constantly giving it haptic feedback from the hands, or it goes off course.

    A rototiller could be compared to a hired hand plowing himself, I guess, but there's way less micromanagement with a hired hand vs a rototiller.

    Kind of like horses and cars. Horses can get you home if you're drunk. Cars can't.

    The proper use of AI agentic tools is like operating heavy machinery. Juniors can really hurt themselves with it but seniors can do a lot of good. The analogy goes further: sometimes you need to get out of the backhoe and dig with smaller tools like jackhammers or just shovels. The jackhammer is like copilot -- a mid-grade power tool -- and Claude code is like the backhoe. Clunky, crude, but can get massive amounts done quickly, if that's what's needed.



    the only good part was the joke about "vibecoding" (shudder what a stupid term) being like a fart and attracting flies... ok investors

    still, this "ai code tools will deprecate real programming" bullshit will one day be laughed at just like how most of us laugh at shitcoin maniacs

    it just takes a lot of people way too long to learn



    Maybe there's a different universe out there, where the code you write is not expected to work, so you can poke the LLM for a whole day to see if it barfs something out.

    I spend much of the day reading and thinking and only a small portion actually writing code, because when I'm typing, I usually have a hypothetical solution that is 99% correct and I'm just bringing it to life. Or I'm refactoring. You can interrupt me at any time and I could give you the complete recipe of what I'm doing.

    Which is why I don't use LLMs, because it's actually twice the work for me. Typing out the specs, then verifying and editing the given result, while I could type the code in the first place. And they suck at prototyping. Sometimes I may want to leave something in the bare state where only one incantation works, because I'm not sure of the design yet, and have a TODO comment, but they go to generate a more complicated code. Which is a pain to refactor later.



    Karpathy coined the term "vibe coding"? His lectures are a bit vibe driven, too. Read Bengio et al. to understand what is going on.

    Did we ever hear any kind of apology from these people that they try to ruin software development and cash in, using the works of others?







    Join us for AI Startup School this June 16-17 in San Francisco!


    Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact



    Search:
    联系我们 contact @ memedata.com