conv.

All stories
AIRunning 5d ↳ Engineer uses LLMs to build interactive simulations for learning complex topics

Domain expertise determines LLM effectiveness; skill required despite democratization

Tech discourse argues that LLMs reward deep knowledge in a field, not just prompting technique, with mathematician Terence Tao exemplifying expert use.

11 items1.4k engagement576 comments Hacker NewsLobsters

Conversation activity · last 7 days peak 4/hr

Peak 4 items in one hour at Aug 3, 5 PM; 11 items over 7 days Aug 3, 5 PM — 4 itemsAug 3, 6 PM — 2 itemsAug 3, 7 PM — 1 itemAug 3, 8 PM — no itemsAug 3, 9 PM — 1 itemAug 3, 10 PM — no itemsAug 3, 11 PM — no itemsAug 4, 12 AM — no itemsAug 4, 1 AM — no itemsAug 4, 2 AM — no itemsAug 4, 3 AM — no itemsAug 4, 4 AM — 1 itemAug 4, 5 AM — 1 itemAug 4, 6 AM — no itemsAug 4, 7 AM — no itemsAug 4, 8 AM — no itemsAug 4, 9 AM — no itemsAug 4, 10 AM — no itemsAug 4, 11 AM — no itemsAug 4, 12 PM — no itemsAug 4, 1 PM — no itemsAug 4, 2 PM — no itemsAug 4, 3 PM — no itemsAug 4, 4 PM — no itemsAug 4, 5 PM — no itemsAug 4, 6 PM — no itemsAug 4, 7 PM — no itemsAug 4, 8 PM — no itemsAug 4, 9 PM — no itemsAug 4, 10 PM — no itemsAug 4, 11 PM — no itemsAug 5, 12 AM — no itemsAug 5, 1 AM — no itemsAug 5, 2 AM — no itemsAug 5, 3 AM — no itemsAug 5, 4 AM — no itemsAug 5, 5 AM — no itemsAug 5, 6 AM — no itemsAug 5, 7 AM — no itemsAug 5, 8 AM — no itemsAug 5, 9 AM — no itemsAug 5, 10 AM — no itemsAug 5, 11 AM — no itemsAug 5, 12 PM — no itemsAug 5, 1 PM — no itemsAug 5, 2 PM — no itemsAug 5, 3 PM — no itemsAug 5, 4 PM — no itemsAug 5, 5 PM — no itemsAug 5, 6 PM — no itemsAug 5, 7 PM — no itemsAug 5, 8 PM — no itemsAug 5, 9 PM — no itemsAug 5, 10 PM — no itemsAug 5, 11 PM — no itemsAug 6, 12 AM — no itemsAug 6, 1 AM — no itemsAug 6, 2 AM — no itemsAug 6, 3 AM — no itemsAug 6, 4 AM — no itemsAug 6, 5 AM — no itemsAug 6, 6 AM — no itemsAug 6, 7 AM — no itemsAug 6, 8 AM — no itemsAug 6, 9 AM — no itemsAug 6, 10 AM — no itemsAug 6, 11 AM — no itemsAug 6, 12 PM — no itemsAug 6, 1 PM — no itemsAug 6, 2 PM — no itemsAug 6, 3 PM — no itemsAug 6, 4 PM — no itemsAug 6, 5 PM — no itemsAug 6, 6 PM — no itemsAug 6, 7 PM — no itemsAug 6, 8 PM — no itemsAug 6, 9 PM — no itemsAug 6, 10 PM — no itemsAug 6, 11 PM — no itemsAug 7, 12 AM — no itemsAug 7, 1 AM — no itemsAug 7, 2 AM — no itemsAug 7, 3 AM — no itemsAug 7, 4 AM — no itemsAug 7, 5 AM — no itemsAug 7, 6 AM — no itemsAug 7, 7 AM — no itemsAug 7, 8 AM — no itemsAug 7, 9 AM — no itemsAug 7, 10 AM — no itemsAug 7, 11 AM — no itemsAug 7, 12 PM — no itemsAug 7, 1 PM — no itemsAug 7, 2 PM — no itemsAug 7, 3 PM — no itemsAug 7, 4 PM — no itemsAug 7, 5 PM — no itemsAug 7, 6 PM — no itemsAug 7, 7 PM — no itemsAug 7, 8 PM — no itemsAug 7, 9 PM — no itemsAug 7, 10 PM — no itemsAug 7, 11 PM — no itemsAug 8, 12 AM — no itemsAug 8, 1 AM — no itemsAug 8, 2 AM — no itemsAug 8, 3 AM — no itemsAug 8, 4 AM — no itemsAug 8, 5 AM — no itemsAug 8, 6 AM — no itemsAug 8, 7 AM — no itemsAug 8, 8 AM — no itemsAug 8, 9 AM — no itemsAug 8, 10 AM — no itemsAug 8, 11 AM — no itemsAug 8, 12 PM — no itemsAug 8, 1 PM — no itemsAug 8, 2 PM — no itemsAug 8, 3 PM — no itemsAug 8, 4 PM — no itemsAug 8, 5 PM — no itemsAug 8, 6 PM — no itemsAug 8, 7 PM — no itemsAug 8, 8 PM — no itemsAug 8, 9 PM — no itemsAug 8, 10 PM — no itemsAug 8, 11 PM — no itemsAug 9, 12 AM — no itemsAug 9, 1 AM — no itemsAug 9, 2 AM — no itemsAug 9, 3 AM — no itemsAug 9, 4 AM — no itemsAug 9, 5 AM — no itemsAug 9, 6 AM — no itemsAug 9, 7 AM — no itemsAug 9, 8 AM — 1 itemAug 9, 9 AM — no itemsAug 9, 10 AM — no itemsAug 9, 11 AM — no itemsAug 9, 12 PM — no itemsAug 9, 1 PM — no itemsAug 9, 2 PM — no itemsAug 9, 3 PM — no itemsAug 9, 4 PM — no itemsAug 9, 5 PM — no itemsAug 9, 6 PM — no itemsAug 9, 7 PM — no itemsAug 9, 8 PM — no itemsAug 9, 9 PM — no itemsAug 9, 10 PM — no itemsAug 9, 11 PM — no itemsAug 10, 12 AM — no itemsAug 10, 1 AM — no itemsAug 10, 2 AM — no itemsAug 10, 3 AM — no itemsAug 10, 4 AM — no itemsAug 10, 5 AM — no itemsAug 10, 6 AM — no itemsAug 10, 7 AM — no itemsAug 10, 8 AM — no itemsAug 10, 9 AM — no items 4 items · 5 PM
Aug 4Aug 5Aug 6Aug 7Aug 8Aug 9Aug 10

Summary, timeline and people extracted by Claude from 11 items across 2 sources · 21h ago. Quotes are verbatim.

Sean Goedecke published an essay arguing that effective LLM use depends primarily on domain expertise, not prompting skill alone, illustrated by mathematician Terence Tao's ChatGPT conversation about the Jacobian Conjecture. The piece sparked discussion on Hacker News about whether LLMs truly democratize technical skills or whether they amplify existing expertise, alongside broader questions about what tasks LLMs should automate in software development.

  • LLM effectiveness is primarily determined by the user's domain expertise, not generic prompting skill, contradicting the idea that LLMs democratize technical work.
  • Expert users signal their knowledge to LLMs, which changes model behavior toward more sophisticated, concise responses suited to practitioners rather than beginners.
  • LLMs function as amplifiers of existing knowledge—extending and directing what users already understand—rather than replacing expertise or enabling complete novices.
  • Design choices about LLM automation in software development carry social and political consequences similar to past technological transitions, and should be examined beyond pure capability.

How it unfolded

  1. Report Peresztegi questions what LLMs should automate

    Adam Peresztegi publishes article examining whether LLMs should automate software development, referencing Langdon Winner's 1980 essay on technology and politics to argue that design choices about LLM automation carry social consequences.

  2. Analysis Prompting compared to medical history-taking

    Commenter tpoacher draws parallels between effective LLM prompting and medical consultation skills, both requiring structured progression from open to specific questions.

  3. Reaction Novice test reveals LLM limitations

    Krisoft describes a test where a non-technical friend attempted to build a web app with LLM help, noting that AI struggled to recognize the user's novice status and provide appropriate scaffolding.

  4. Reaction Demand for formal study of prompting skill

    Commenter Austiiiiii notes the observation matches personal experience but calls for formal research, expressing uncertainty about whether the effect is real or confirmation bias.

  5. Analysis LLMs as amplifying mirrors

    HN commenter abixb characterizes LLMs as 'amplifying mirrors' that extend existing knowledge rather than replace it, distinguishing between those who use LLMs as mind extensions versus replacements.

  6. Reaction HN discussion begins on signaling expertise

    Commenters on Hacker News report that explicitly signaling domain expertise to LLMs changes model behavior, with one user noting that declaring biblical scholarship background alters ChatGPT's responses.

  7. Report Goedecke publishes LLM expertise essay

    Sean Goedecke's article 'LLMs reward expertise' argues that domain knowledge is the most important skill in prompting, using Terence Tao's ChatGPT conversation about the Jacobian Conjecture as primary evidence.

What people are saying verbatim

“The most important skill in prompting is expertise in the domain you're prompting for.”

Sean Goedecke · seangoedecke.com article · Aug 2

“This is not the same ChatGPT I talk to! I couldn't get to where Tao gets, even with unlimited tokens to burn.”

Sean Goedecke · seangoedecke.com article · Aug 2

“By signalling expertise, Tao shunts the model into 'talking-to-mathematicians' mode, not 'explaining-to-amateurs' mode”

Sean Goedecke · seangoedecke.com article · Aug 2

“I do find that 'signalling expertise' is important... That changes things significantly.”

sramsay · Hacker News comment · Aug 3, 5:37 PM

“People who (carefully) use it as an extension of their own mind and senses will very likely thrive, and those who use it as a replacement for their minds and their senses will struggle.”

abixb · Hacker News comment · Aug 3, 6:32 PM

“By far the greatest latitude of choice exists the very first time a particular instrument, system, or technique is introduced.”

Adam Peresztegi (quoting Langdon Winner) · adamperesztegi.com article · Aug 8

Voices from the web unedited

  • I did a test a few months ago. A friend of mine wanted to develop what i understood to be a simple single page web app. But since she didn’t have any software engineering experience she asked me to help. Around that time everyone was talking about how literally anyone can develop software with LLMs i asked her if she could give it a try first, and…

    krisoftHacker News6d agoview on Hacker News ↗

  • Both the article and some of the discussions here share a lot of commonalities with doctors taking a medical history. There is a certain skill in guiding the conversation towards useful outputs, while not dictating the exact outputs to a patient who is eager to please with their responses.E.g., medical history taking protocol always says to start…

    tpoacherHacker News6d agoview on Hacker News ↗

  • The amplifying mirror analogy works best here. LLMs are ultimately a reflection of your own interactions with its weights, the tone you use, the structure with which you construct your prompt, aspects of an issue you tend to focus on, your breadth of vocabulary and world knowledge and whatnot.People who (carefully) use it as an extension of their…

    abixbHacker News6d agoview on Hacker News ↗

  • This is something that really needs to be formally studied.I'm inclined to say that this matches my own experience, but I can't rule out confirmation bias on my part.As a meticulous person generally looking for a very specific code outcome, I prompt in a way intended to get exactly the thing I have in mind, and my results reflect that. But on the…

    AustiiiiiiHacker News6d agoview on Hacker News ↗

  • > Of course both are useful, but I’d rather have familiarity with the codebase than a deep general understanding of software systems.In my experience, getting that familiarity with a particular codebase in a way that isn't surface-level has always been a hands-on process. E.g. just because I know many general things about software, I need to know…

    dbalateroHacker News6d agoview on Hacker News ↗

  • The counterexample of the Dinitz-Garg-Goemans conjecture was basically just "keep going" and finally "enough of partial results. now finish with a complete unconditional counterexample" lol

    porphyraHacker News6d agoview on Hacker News ↗

  • This reminded me of Gaussian Processes. You start out with n-dimensional unconstrained (but strongly correlated) gaussians. As soon as constraints (data) are added (mathematically it's called conditioning), the thing goes more and more into shape.Prompting feels a lot like this conditioning phase to me. You start with an LLM in unconstrained mode…

    cgufusHacker News6d agoview on Hacker News ↗

  • I do find that "signalling expertise" is important. "I have a significant background in biblical scholarship. You can assume I've read the most important works in NT studies in particular. Do not translate Greek, Latin, Hebrew, or Syriac. Now, I would like to know . . ." That changes things significantly. So does telling it you have 20+ years of…

    sramsayHacker News6d agoview on Hacker News ↗