Item — HN Reader

> Input Risk. An LLM does not challenge a prompt which is leading...

(Emphasis mine)

This has been the biggest pain point for me, and the frustrating part is that you might not even realize you're leading it a particular way at all. I mean it makes sense with how LLMs work, but a single word used in a vague enough way is enough to skew the results in a bad direction, sometimes contrary to what you actually wanted to do, which can lead you down rabbit holes of wrongness. By the time you realize, you're deep in the sludge of haphazardly thrown-together code that sorta kinda barely works. Almost like human language is very vague and non-specific, which is why we invented formal languages with rules that allow for preciseness in the first place...

Anecdotally, I've felt my skills quickly regressing because of AI tooling. I had a moment where I'd reach out to it for every small task from laziness, but when I took a real step back I started realizing I'm not really even saving myself all that much time, and even worse is that I'm tiring myself out way quicker because I was reading through dozens or hundreds of lines of code, thinking about how the AI got it wrong, correcting it etc. I haven't measured, but I feel like in grand totality, I've wasted much more time than I potentially saved with AI tooling.

I think the true problem is that AI is genuinely useful for many tasks, but there are 2 camps of people using it. There are the people using it for complex tasks where small mistakes quickly add up, and then the other camp (in my experience mostly the managerial types) see it shit out 200 lines of code they don't understand, and in their mind this translates to a finished product because the TODO app that barely works is good enough for an "MVP" that they can point to and say "See, it can generate this, that means it can also do your job just as easily!".

To intercept the usual comments that are no doubt going to come flooding in about me using it wrong or trying the wrong model or whatever, please read through my old comment [1] for more context on my experience with these tools.

[1] https://news.ycombinator.com/item?id=44055448

8 comments

From my experience so far it is helpful for me to get another opinion on how to solve a problem—and I do the work in the end. Or, I am extremely specific, and give it a relatively small problem to solve, and it solves it—writes the code for me—and then I code review it, and make changes to uphold my standards.

In other words, AI is my assistant, but it is MY responsibility to turn up quality, maintainable work.

However, to put things in perspective for the masses: just consider the humble calculator. It has ruined people’s ability to do mental math. AI is going to do that for writing and communication skills, problem solving skills, etc.

> From my experience so far it is helpful for me to get another opinion on how to solve a problem—and I do the work in the end.

I agree fully, I use it as a bouncing off point these days to verify ideas mostly.

The problem is, and I'm sure I'm not alone in this, management is breathing down my neck to use AI for fucking everything. Write the PR with AI, write the commit message with AI, write the code, the tests, use YOUR AI to parse MY AI's email that I didn't bother proofreading and has 4 logical inconsistencies in 1 sentence. Oh this simple feature that can easily be done for cheaper, quicker and easier without AI? Throw an AI at it! We need to sell AI! "You'll be left in the dust if you don't adopt AI now!"

It comes back to my point about there being 2 camps. The one camp actually uses AI and can see their strengths & weaknesses clear as day and realizes it's not a panacea to be used for literally everything, the other is jumping headfirst into every piece of marketing slop they come across and buying into the false realities the AI companies are selling them on.

> but a single word used in a vague enough way is enough to skew the results in a bad direction

I'm glad I'm not the only one who feels this way. It seems like these models latch on to a particular keyword somewhere in my prompt chain and throw traditional logic out the window as they try to push me down more niche paths that don't even really solve the original problem. Which just leads to higher levels of frustration and unhappiness for the human involved.

> Anecdotally, I've felt my skills quickly regressing because of AI tooling

To combat this, I've been trying to use AI to solve problems that I normally would with StackOverflow results: for small, bite-sized and clearly-defined tasks. Instead of searching "how to do X?", I now ask the model the same question and use its answer as a guide to solving the problem instead of a canonical answer.

Definitely share your feeling that people move the goalposts from "AI can do it" to "well it would have been able to do it if you used model o2.7 in an IDE with RAG and also told it how to do it in the prompt" ...ok, at some point it's less value for the effort than writing the code myself, thanks

That said, AI does make some things easier today, like if you have an example to use for "make me a page like this but with data from x instead of y". Often it's faster than searching documentation, even with the caveat that it might hallucinate. And ofc it will probably improve over time.

The particular improvement I'd like to see is (along with in general doing things right) finding the simplest solution without constantly having to be told to do so. My experience is the biggest drawback to letting chatgpt/claude/etc loose is quickly churning out a bunch of garbage, never stopping to say this will be too complex to do anything with in the future. TFA claims only humans can resist entropy by understanding the overall design; again idk if that will never improve but it feels like the big problem right now.

What I've been doing when I want to avoid this "unexpected leading", is to tell the LLM to "Ask me 3 rounds of 5 clarifying questions each, first.". The first round usually exposes the main assumptions it's making, and from there we narrow down and clarify things.

I've read you comment about all the things you tried, and it seems you have much broader experience with LLMs than I do. But I didn't see this technique mentioned, so leaving this here in case it helps someone else :).

People have similar biases, it's just a lot easier to test an LLM for biases than it is to test humans so we are more aware of LLM biases.

I consider this hype the preparatory phase for the era where we rediscover the laws of quality control.

A GREAT example is good old Coke vs Pepsi.

Almost all of ai hype seems to be on the backs of no coders or people who make toys as a career (if they even have one)

Nonono you've got it all wrong you clearly need to try [model name] with [coding product], it 100x'd my productivity!