本文由 AI 分析生成
建立時間： 2026-03-28 來源： https://news.ycombinator.com/item?id=42938409
Summary
Hacker News discussion on Sean Goedecke’s article about LLM use as a staff engineer. The thread captures the full spectrum of developer opinion: strong skeptics (LLMs produce unreviewable garbage and teach bad practices), pragmatic adopters (useful for specific tasks with developed intuition), and career pragmatists (anti-LLM positions may be career-limiting).
關於 Sean Goedecke 使用 LLM 的 HN 討論。討論涵蓋了開發者意見的全貌：強烈懷疑者（LLM 生成難以審查的垃圾並教授不良實踐）、務實採用者（對特定任務有用，需要培養直覺），以及職業務實主義者（反 LLM 立場可能影響職業發展）。
Key Points
- Skeptic view: LLMs cause “Faustian deal” — learn terrible practices, rely on boilerplate/non-deterministic outputs, hurt code review throughput
- Pragmatist view: LLM utility requires developed intuition (“muscle memory for when it’ll work”); teachable to most engineers in ~1 week
- Career pragmatist view: executives want AI adoption; being anti-LLM may be “detrimental for your career” regardless of technical merit
- Core tension: non-deterministic outputs make LLM-assisted code fundamentally different from traditional code generation (compilers, templates)
- Meta-observation: “using them isn’t obvious” — the biggest misconception is that LLMs are easy to use without serious effort
Insights
The “lotto numbers” critique is the most philosophically interesting: success stories with LLMs may be like startup success stories — the conditions that made it work for one person aren’t generalizable because they depend on specific non-reproducible factors (the exact model, the exact problem type, the developer’s specific existing knowledge). This is different from learning math or programming where mastery does generalize. The counter is that shared patterns (code generation vs. architecture decisions, throwaway code vs. production code) do generalize enough to create real productivity differences.
Connections
Raw Excerpt
Using them isn’t “obvious”. The single biggest misconception about LLMs is that they are easy to use and you don’t need to put a serious amount of effort into learning how to best apply them.