Thread

Loading the complete thread in the background. This saved snapshot is available now. Refresh

1. blktiger · 1790799049 · focus · HN ↗

   I have to disagree a bit here, people (including me sometimes) can take the lazy way out and just tell you what an agent said in response to a question which we shouldn't do. At that point it's little better than a let me google that for you link. However, while I may be passing along what I found using an agent, but usually I still had to do the work of prompting it, asking it to lookup information for me, and otherwise filtering the agent down to the relevant information. There is a good chance that I'm doing that because of my specific expertise in the subject area which is something outside of what you could have found by just asking an agent yourself.

   1. mbil · 1790801401 · focus · HN ↗

      Yes, a subject matter expert who uses their domain experience to word the prompt correctly, incorporate tacit knowledge and relevant sources, ask follow-up questions, and judge the LLM-generated responses for relevance and correctness is not taking a shortcut. They're using a tool to provide a better and faster answer than they would have been able to otherwise. It's a matter of how the tool is used and maybe the optics around how the response is communicated, but relaying Claude's final answer is not inherently the wrong thing to do.

      1. JohnFen · 1790802671 · focus · HN ↗

         > relaying Claude's final answer is not inherently the wrong thing to do.

         But it's very poor communication, and (to a whole lot of people) comes off as dismissive and unhelpful. It's better to hear from the subject-matter expert than from genAI directly.

      2. jmoggr · 1790803360 · focus · HN ↗

         The problem is that I don't know if that work has been done, and in the cases where it has not been done it is expensive for me to trust the output. In my limited experience the work is often not been done, and given the cost to me of consuming low effort AI slop, leads me to discount anything looking like AI output.

         Unfortunately its not about the tool, its about trust. It takes little to destroy that trust, and without it we are denied the possible benefits from the tool.

      3. otabdeveloper4 · 1790837079 · focus · HN ↗

         > ...a subject matter expert who uses their domain experience to word the prompt correctly, incorporate tacit knowledge and relevant sources, ask follow-up questions, and judge the LLM-generated responses for relevance and correctness

         ...doesn't need an LLM and would be better served by decent cross-indexed reference documents.

2. MisterKent · 1790799060 · focus · HN ↗

   I'm reasonably opposed to LLMs, mostly from a human perspective. They reduce us to our output without factoring in any of the actual thoughts and experiences that go into that work, it's literally not in their training data.

   It's unfortunate that we live in a world that has decided to push in this direction.

   I have started a work project that is 100% vibe engineered, as an experiment to see where it would land. While it seems to be working, I can see that it has made so many mistakes along the way that it is still unclear if we can dig ourselves to production ready.

   I was at an art museum recently, and was looking at a marble statue. And I thought to myself: this artist probably spent more time sculpting the big toe than anyone has spent appreciating the entire statue.

   That speed is what we used to have in software engineering. So much careful craftsmanship, no line just yeeted out. And now we're just heading into a direction with no clear end, just prayers that bigger and better models will lift us forward.

   What's the point of being human if we're so gung ho to remove that which separates us from other animals (our intellect)?

   1. jolt42 · 1790800198 · focus · HN ↗

      You still have your appreciation for music and laughter.

      1. MisterKent · 1790802397 · focus · HN ↗

         Yeah, I have no interest in becoming a pet. Especially not as a species.
   2. manmal · 1790803346 · focus · HN ↗

      > That speed is what we used to have in software engineering. So much careful craftsmanship, no line just yeeted out.

      That reads like nostalgia. What you describe sounds like the code I brought forth on a sunny Sunday afternoon, but certainly not on a hectic work day under deadline stress. I had Stackoverflow open _all the time_, and started googling at the earliest inconvenience.

3. vb-8448 · 1790799116 · focus · HN ↗

   It's "I saw it on Google" on steroids.
4. SuperV1234 · 1790799208 · focus · HN ↗

   Human slop.

   If someone answers your question with "Claude says..." it's a clear indication that you should've asked the question to an LLM yourself before bothering a human.

   This is just the new version of LMGTFY.

   > AI is the reason our software stack is overengineered to the 10th degree

   Complete BS.

   1. ApolloFortyNine · 1790800992 · focus · HN ↗

      There's definitely times where the question is 'hard' enough (usually having to do with something so business specific the LLM won't have a great answer) where a true human answer is needed, but I do regularly still get asked questions of 'why is x failing' with a datadog link. And if you shove the exact logs they've already found that contain a stack trace, the LLM instantly tells you exactly what was wrong.

      1. SuperV1234 · 1790808643 · focus · HN ↗

         This is my experience as well. Some of my coworkers ask questions in chat that could be trivially solved by querying Claude. If I were tasked with answering that question, the responsible thing to do would be to query Claude rather than risking a manually-written incorrect answer off the top of my head. So yes, I will reply with "Claude says..."
5. thih9 · 1790799358 · focus · HN ↗

   This is pre-AI speak. Post-AI assumes we all maintain multiple concurrent chats and use that to search for information. “Claude says” becomes the new “I googled this”.

   I started writing this as satire and I don’t like how realistic this gets.

   I still don’t like it, if only because the power imbalance seems somehow worse than with Google.

6. atomicnumber3 · 1790799365 · focus · HN ↗

   I tire of all of this discourse at this point because people who don't need to hear it are reading it and nodding, and the people who DO need to hear it are busy making everything worse, earning leadership accolades (or are the ones handing them out), and are completely closed to any criticism or input.

   Anyone with intellectual curiosity and any sense of subtlety or balance is operating in a separate reality from people who are so AI-pilled and very proud of it and there's basically no possible cross talk (usually because the AI-pilled people are the ones who own all the shares).

   I say this as someone who really likes LLMs and what roles they'll play in software engineering in the future (and already are, right now).

   I can't wait until the AI craze is in the same spot that the crypto craze is currently. People are still using and building with crypto, but the sale of ape PNGs has mostly stopped, and now if I have a problem that a blockchain could solve, I could actually talk about it instead of accidentally picking up a battle standard in the middle of a trench war.

   1. ncr100 · 1790801971 · focus · HN ↗

      [delayed]
   2. pwillia7 · 1790802766 · focus · HN ↗

      > accidentally picking up a battle standard in the middle of a trench war.

      ...nice

7. kittikitti · 1790799531 · focus · HN ↗

   It's become so frustrating when literally everyone is just using Claude for everything. Simple messages and updates are fed through Claude. It's come to the point where any human language in corporate is punished.

   1. chrisjj · 1790802589 · focus · HN ↗

      [delayed]
8. cush · 1790800614 · focus · HN ↗

   Disclosing that you got some info from an AI is way better than not. We’re all using AI in some capacity just be transparent about it. If you didn’t read the findings then say it. As always there’s nuance and it depends
9. throw310822 · 1790800747 · focus · HN ↗

   No problem, the next time you have an issue or ask if anyone has any idea, I'll just ignore you and mind my business, confident that you can do the prompting yourself. Problem solved.

   1. JohnFen · 1790840668 · focus · HN ↗

      Honestly, I would very much prefer that to people just copypasting genAI reponses.
10. ectoloph · 1790800863 · focus · HN ↗

    We live in an interesting time.

    I have been in meetings where the conversation has been 'my Claude has got feedback', 'great, I'll get my Claude to take a look'.

    When do you just cut out the human middleware in that process?

    1. cush · 1790801073 · focus · HN ↗

       We’re currently entering that time and people are quickly learning that empirical evidence is the new output. It’s the difference between “I think the bug is here” vs “I fixed the bug and here’s the evidence I gathered from user telemetry”
11. jchw · 1790801316 · focus · HN ↗

    The one thing I do find extremely frustrating is how there can be some ridiculously painful obstacle that hits you in the face all the time when trying to ship code, and everyone is seemingly sympathetic, but it can't be changed for dubious compliance reasons, or maybe because if the barriers were lowered it would threaten the quality control for the code. Fair enough.

    Then, of course, all of those speedbumps are demolished in record time, only for the AI agents and not for the humans, regardless of what caveats remain unsolved, with little regard for whether or not it meets compliance standards.

12. aranelsurion · 1790801664 · focus · HN ↗

    It took maybe a decade to teach people to not use "@here" on Slack, and avg. response time of an email isn't 10 minutes. Technology comes out, many people fumble through for years, until finally some etiquette is formed. Happens all the time.

    I don't remember the last time someone dropped me the first link out of a Google search, telling me they haven't read themselves and have no idea about its contents. Yet I see it happen in various 1-1 and public settings with Claude all the time.

    In those discussions people often don't use this as a "Here let me Claude it for you" kind of protest, but genuinely they mean "I copy pasted this topic to Claude, and Claude said this. I'm meat-proxying to you, and absolutely haven't invested a single microsecond of my time on it". Obviously the expectation is you'd invest the effort they haven't.

    My strategy and hope, is to give it a decade. It will probably get better.

13. vinni2 · 1790802212 · focus · HN ↗

    Wasn’t it also like google says?
14. thefourthchime · 1790802546 · focus · HN ↗

    I find it a little irritating when somebody literally copies and pastes a bot response without editing it at all. Not because it's wrong, but because it's full of cruft and slop. The core information is there, but it's surrounded by junk the author probably doesn't understand and hopes I will.

    As a software engineer talking to other software engineers, I'm the controller of my domain knowledge. But my domain knowledge is now subjugated to Claude, and it tells me what it did. When someone else needs to know what I did, I have to ask Claude what it did. Then I send it over the fence and the other engineer does the same.

    I don't think this is inherently bad. It's sort of like two managers talking to each other and having to talk to their employees about what's actually going on. What's bad is when the managers themselves don't understand it and just start blindly copying and pasting things. That's where you get into trouble.

    1. LoganDark · 1790803294 · focus · HN ↗

       If someone shuttles my question to AI the least they can do is extract the meaningful information themselves and communicate just that to me.
15. scottyah · 1790804645 · focus · HN ↗

    I disagree completely, citing your source is an amazing habit and I wish more people did it. "Webster says" vs "My mamaw told me" makes a big difference, and to all those like the author who think that drudging through manuals on specific topics will pay off in 20yrs, I really think you should look into how human memory works. One of my favorite quotes is "That guy has forgotten more about SSL than I've ever learned, and he still made a typo that broke the internet."

    So please, cite your sources where relevant and don't waste people's time with your fuzzy human memories.

    1. queenkjuul · 1790814952 · focus · HN ↗

       Yeah i wish Claude cited sources

       1. scottyah · 1790916177 · focus · HN ↗

          Google's web search AI is pretty great at it!
16. Tadpole9181 · 1790806247 · focus · HN ↗

    I feel like the author has been really lucky with their coworkers, or is not in a leadership position.

    My entire professional life I have dealt with people who refuse to do basic research. They won't diagnose, they won't investigate, they won't RTFM. Instead, they'll ask me questions as the very first step.

    Just like I would forward them information like "the docs say...", I'm going to start by forwarding information like "Claude says..." so as to not waste my own time until they give me a specific reason they're asking me. Odds are they're actually just being lazy / intellectually incurious and haven't even Googled it or asked Claude.

    Now if someone sends me a well thought out question with what specifically they're having trouble with, I'm more than happy to actually dig in. But odds are, questions like that will include research and auditing, which I'm going to do with Claude. I'll double-check its work, but I still want to explain what information is coming from me and what information is coming from it. They can use that to evaluate the level of trust they have in that information, instead of me making the decision for them.

    I am empathetic to the author's point, though. It's frustrating to get a taste of my own medicine on this, but I often take that as a sign that I'm doing the thing I hate and should have asked a better question. If I reflect and still believe otherwise, I message them politely about it. And if it repeatedly happens, I escalate to our manager - like I did before Claude existed.

17. deterministic · 1790820341 · focus · HN ↗

    This post doesn’t really add anything new. These points about AI and critical thinking have been made many times before.

Unofficial Hacker News client; not affiliated with Y Combinator.