HackerNews中文版

我在播客中向四位在职工程师提出了完全相同的问题：一位谷歌开发者关系倡导者（斯德哥尔摩）、一位高级软件工程师/顾问（巴黎）、一位英伟达深度学习研究所讲师（摩洛哥）以及一位 IBM 基础设施工程师（都柏林）。以下是他们实际的回答： * 高级软件工程师说：“大型语言模型就像婴儿。如果你不理解其背后的架构，就无法跟上。” * 谷歌的倡导者略有不同意见：“编写代码已经变得像商品一样，就像自动化后的汽车制造。问题不在于是否要学习编程，而在于你为什么想学。” * IBM 的基础设施工程师给出了最具操作性的建议：“不要把人工智能当作代笔。切勿提交你无法解释的代码。把它当作导师，而不是你独立思考的替代品。” 还有更多内容。我们还对一段硅谷高管告诉大学毕业生“人工智能是下一场工业革命”并遭到人群嘘声的视频片段做出了反应。一个更令人警醒的部分是：按照目前的用量水平，Anthropic 可能会在重度 Claude 订阅用户上亏损。目前没有一家推理公司是盈利的。在谷歌工作的、负责大规模 LLM 推理的倡导者直言不讳地说：如果我们未来五年内无法在推理效率上取得显著的改进，整个生态系统将变得难以负担。而英伟达则有更多关于人工智能成本效益的细节可以分享。

查看原文

I asked 4 working engineers this exact question on my podcast: a Google Developer Advocate (Stockholm), a Senior Software Engineer/consultant (Paris), an NVIDIA Deep Learning Institute Instructor (Morocco), and an Infrastructure Engineer at IBM (Dublin).Here's what they actually said:- The Senior Software Engineer said: "LLMs are babies. If you don't understand the architecture behind everything, you won't be able to follow."- The Google advocate pushed back slightly: "Writing code has become a commodity, like car manufacturing after automation. The question isn't whether to learn to code, it's why you want to."- The IBM infrastructure engineer had the most actionable take: "Don't treat AI as a ghostwriter. Never commit code you can't explain. Use it as a tutor, not a replacement for your own thinking."And many moreWe also reacted to a clip of a Silicon Valley exec telling university graduates that "AI is the next industrial revolution" and getting booed by the crowd.One of the more sobering parts: at current usage levels, Anthropic is likely losing money on heavy Claude subscribers. No inference company is profitable today. The Google advocate, who works on LLM inference at scale, put it directly: if we don't get significant inference efficiency improvements in the next five years, this entire ecosystem becomes unaffordable. And the NIVIDA have more details to talk about AI cost effectiveness

Ask HN：软件工程对新生来说仍然是好的职业选择吗？