What's this about?
Sean Goedecke argues LLMs are an amplifier, not a replacement: the more you already know, the more you get out of them — experts extract 10× value while novices get slop, because value comes from asking precisely the right thing.
What exactly does "reward expertise" mean?
The tool's output quality scales with the quality of your questions and your ability to judge answers. Like a very fast junior employee: a good team lead gets great work out of them, someone who can't specify or review gets confident garbage.
What's the community saying?
HN 502 points, 221 comments.
- Broad agreement via the team-lead analogy: "you're basically playing team lead to the LLM's junior dev."
- The Anthropic-mathematician example gets dissected: expert prompts differ less by magic words, more by knowing what to ask for (and Tao used chats to build intuition, not outsource proofs).
- A meta-observation: every commentator concludes LLMs empower *their own* group — evidence we're all generalizing from ourselves.
- Extension to images/video: prompt a generator without photography knowledge and you get slop — same law, different medium.