Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Don't be a meat proxy

gruhn.me

Post reply on HN
Print view View original

481–490 of 786 posts

Re: Don't be a meat proxy

#481

Post by TacticalCoder »

plutokras wrote: post #264
gregsadetsky wrote: post #243

On social media, I saw the much more vulgar “Learned engineering just to become the condom between Claude Code and prod” And that (re)framing helped as well to think about the “what are we even (left) doing” as an industry

With the recent wave of posts about Claude rewriting entire codebases from Zig/Electron to Rust/Swift, I can't help but question the prompter's actual confidence in the final result. Gaining a thorough understanding of the generated output still demands an enormous amount of effort. I suspect the reality is that most people just chain a few review agents together, and if they output "LGTM", the code gets committed an…

> The foundational knowledge is lost ...

Not yet. The world still pretty much runs on systems not written by LLMs. And humans wrote all of that. LLMs runs on OSes and hardware created by "not LLMS".

And seen the "quality" of the output of LLMs (and I pay three AI subscriptions), I don't think we're getting rid of human-produced code, OSes, hardware, anytime soon.

Re: Don't be a meat proxy

#482

Post by felipeerias »

One way to prevent obvious AI language from sneaking in text destined for other human beings is to ask the model to produce ASD-STE100 Simplified Technical English bullet points.

This will result in a list of sentences that are clear and explanatory, easier to double-check, and convenient for the user to rewrite into a more readable format with a human voice.

Re: Don't be a meat proxy

#483

Post by erikerikson »

deepsummer wrote: post #39
Maxious wrote: post #21

Earlier quoted context omitted.

So if you used claude you would preface with "Claude said: "

Sometimes I do that. The problem is that it makes them look bad.

Whose responsibility is that?

To be fair those are consummate and thoughtful politics. The industry is rife with those but that was part of why in leaving it.

Re: Don't be a meat proxy

#484

Post by stellamariesays »

eddythompson80 wrote: post #8

I deal with this all day long at work and it’s exhausting. People almost acting like no one has thought of it “I asked Claude what happened, and it spit out this 300 line response. Can you read it for me and see if it’s right?” What kills me is you might expect this from a busy high level manager that doesn’t really understand the technical details and they just point the AI to an error they got. They don’t know how…

[flagged]

Re: Don't be a meat proxy

#485

Post by bigfishrunning »

xyzelement wrote: post #445

// Claude said: [giant response verbatim] I am lucky that I work with great teams and so when someone says "this is how Claude (or Gemini) summaries the situation" it does mean "I read it and it's right-enough to be helpful, so I am passing it along without editing" Obviously for this to work you need (1) a team that's smart and mature enough to own the AI output it puts forward and (2) shared understanding that peop…

So the answer was "What's going on with this account" was 10 pages of homework, with at least 1 mistake to correct? That doesn't seem like an improvement over a conversation.

Re: Don't be a meat proxy

#486

Post by washbasin »

Great security talk about how humans aren't setup to deal with LLMs as tools.

I love the bullet point(s) : we don't trust [aircraft] pilots to catch automations mistakes (e.g. autopilot), but we are expecting humans to catch LLM mistakes.

Artificial Intelligence Natural Stupidity - Brandon Sherman

https://www.youtube.com/watch?v=7bx9UfDHf_k

Re: Don't be a meat proxy

#488

Post by WillMorr »

eddythompson80 wrote: post #8

I deal with this all day long at work and it’s exhausting. People almost acting like no one has thought of it “I asked Claude what happened, and it spit out this 300 line response. Can you read it for me and see if it’s right?” What kills me is you might expect this from a busy high level manager that doesn’t really understand the technical details and they just point the AI to an error they got. They don’t know how…

The thing I think is so frustrating about this is it's the absolute worst case middle ground between two much more productive experiences. Either solve the problem yourself by understanding the code or just give me the full agent context so I can just use a chatbot too.

For high stakes debugging efforts I'll dig in with Claude, have it do a bunch of testing and give me a writeup, and then review it to extract any useful findings. When discussing with other people, I give them the 2 sentences I'm confident in and then link them to the giant doc so they can review it with their Claude.

IMO once you're working with agents, your whole job is context management. I have a little web server running my Claude markdown docs. When someone sends me a Claude snippet with some partial information I just have them prompt their bot to upload a thorough context dump so I actually know what they did. If someone has had an agent look at a problem they haven't thought about yet, I'd rather just get the info from the bot directly.

Re: Don't be a meat proxy

#489

Post by Havoc »

72deluxe wrote: post #447
Havoc wrote: post #192

The tricky bit is politely conveying this to people that lets call it don't think about things too deeply. They genuinely think they're helping by shoveling 5 pages of LLM output they didn't even bother reading into teams...

But if some people lack the understanding ability to even realise that they lack understanding, how do you tell them? Nightmare.

You don't. Pretty sure the last instance where this happened the other person thought I was just anti-AI when I voiced my displeasure. I don't think the actual problem registered

How others in office context attempt to use it is starting to annoy in general. Had a case today where someone wanted to use copilot to classify something that is situation/company specific & out of distribution. There is just no understanding beyond "its a magic answer box"

Re: Don't be a meat proxy

#490

Post by freehorse »

throw310822 wrote: post #448
freehorse wrote: post #263

Earlier quoted context omitted.

You took all this time to reformulate the question, add the relevant implicit context, and do back and forth BUT you did not have time to reformulate the answer to something human readable that YOU own and can take responsibility for? This does not make sense to me. Nobody said not to consult an llm while formulating an answer. But if you give an answer, own it. Don't throw slop grenades.

Why would I reformulate what Claude already explained in a way that I find satisfactory? Only to give you the impression that I came up with it myself? > if you give an answer, own it. Don't throw slop grenades. If I send you some text copied and pasted from Claude, it means that I have read it and I think it's worth for you reading it. In this sense I'm already owning it. The problem is not the pasted answer, it's y…

I would at least expect to share your own take on it, or what you think is interesting or important in sharing an answer. If I am sent an llm copypasta I have no idea of the amount of intentionality or the amount of effort is being put into this. I have been sent many times "I got this from claude, did not have time to look at it closely, maybe you find it interesting". No thanks, I can and probably already have tried with claude already (did not say exactly that). This is the kind of slop grenade that is basically noise, and when you slop grenading me without telling me what you think yourself I have no idea if it is worth my time reading it or not.

I feel like these things are pretty trivially obvious. When there is noise, you have to make sure your signal is not taken for noise, if we assume that your intention is to produce signal in the first place.

Post reply on HN