Does your work not depend on existing code bases, product architectures and extensive domain contexts the LLM knows nothing about?

Every thread like this over the past year or so has had comments similar to yours, and it always remains quite vague, or when examples are given, it’s about self-contained tasks that require little contextual knowledge and are confined to widely publicly documented technologies.

What exactly floored your colleague at Microsoft?

Solid questions and comments, layer8.

I notice that the person you replied to has not replied to you yet.

It may be that they will, of course.

But your points are good.

Even more negativity towards this comment of mine, than to some of my other ones, as shown by the greater downvotes.

Jeepers creepers! :)

Not-so-subtly mocking the top-level for not replying "yet", when they replied almost immediately after with a video of the relevant workflow, was not a move that made you look smart or nice.

Hey.

These are the four simple lines that I wrote above:

>Solid questions and comments, layer8.

>I notice that the person you replied to has not replied to you yet.

>It may be that they will, of course.

>But your points are good.

(Italics mine, and they were not in my original comment.)

You, above:

>when they replied almost immediately after with a video of the relevant workflow,

I did check the time intervals between the top level comment and layer8's comment, before my first reply. It is over an hour now, so I cannot see the exact times in minutes any more, but IIRC, there was a fairly long gap (in minutes). And I also think I noticed that the top level person did reply to someone else, but not to layer8, by the time I wrote my comment.

So I don't see anything wrong in what I said. I even said that they may reply later.

You consider that to be:

>"Not-so-subtly mocking"?

Jeez. I think you are wrong.

Then I have nothing further to say to you, except this:

>was not a move that made you look smart or nice.

Trying to look smart or nice is not the goal in online discussions. At least, I don't think so. You appear to think that. The goal (to me) is to say what you think, otherwise, why write at all? I could just get an LLM to write all my comments, and not care about its hallucinations.

I don't try to be smart or nice, nor the reverse. I just put my considered thoughts out there, just like anyone else. Obviously I can be right or wrong, just like anyone else can be. And some points can be subjective, so cannot be said to be definitely either right or wrong.

You were being passive aggressive and adding nothing to the discussion. The fact that you weren't wrong doesn't make any difference.

>You were being passive aggressive and adding nothing to the discussion.

What arrant nonsense!

Don't accuse people falsely or without substantiating your accusation.

My comment being referred to above, consisted of four simple short lines.

I am not going to bother to paste them here again for a third time.

Those HN readers who wish to do so, can go the few needed levels up this thread and check for themselves:

Nothing I said in those four lines can be interpreted as being passive-aggressive.

I say that you do not know what the heck you are talking about, if you claim that I was passive-aggressive in that comment.

Placing statements on separate lines does come off as somewhat aggressive. It may be subtle, but I got a similar impression.

Your comment is so hilarious that I need some time to analyse its implications and think of a suitable reply.

Mañana.

If a comment is not at least one of smart or nice, it's a waste of space and attention. That may not be your purpose, but don't act shocked when people respond with negativity.

>when they replied almost immediately after with a video of the relevant workflow

Wow. Such wrong claims.

I had already replied to you in a sibling comment, refuting your points, but will give one more proof (not that I really need to):

_acco, the top level commenter relevant to this discussion, commented at some time, say x.

layer8 commented, replying to _acco, 7 hours ago (as can be seen on the page at the time of my writing this comment, i.e. right now).

I then replied to layer8, 6 hours ago.

_acco replied back to layer8 5 hours ago.

All this is visible right now on the page; and if people check it a few hours later, the relative time deltas will remain the same, obviously. (But not if they check after 24 hours, in which case all comments will show as one day ago.)

So there was a 1 hour gap between layer8's comment and mine, and a 2 hour gap between layer8's comment and _acco's reply.

If you think 2 hours is the same as "almost immediately", as you said above, I have nothing more to say to you, except that our perceptions of time are highly different.

2 hours in a discussion forum, where the discussion spans days or sometimes weeks is certainly an ”almost immediate” response.

Perception of time is subjective.

Err ... you realize that your argument just shot itself in both left feet?

I just love it (not!, but it happens every now and then) when HN users display ignorance of basic logic, on a site for techies, yet.

I could use your own argument against you:

>is certainly an ”almost immediate” response

>Perception of time is subjective.

So, you use "certainly", and then "subjective", just after it, in the same argument about the same topic?

Brilliant. Do you not realize that by your last sentence above, you make my own case for me?

I meant immediately after your reply. At the time I posted, your and acco_'s replies to layer8 both showed as "3 hours" ago. Now they both show as "13 hours ago". Really, I'm being generous in assuming they didn't reply before you.

Ed: ah, since the time I wrote this comment, your respective comments are now at 14 and 13 hours. Congrats on your <1hr lead.

By god, andrewflnr. Very "nice". /s. See point 4 below.

You just showed that you are inaccurate, pompous and fake, all of that, in one single comment of yours, above. How? I'll tell you:

1. inaccurate: That commenter's username (the one who started this subthread) is _acco, not acco_ as you wrote above.

Check that in their comment, or here:

https://news.ycombinator.com/user?id=_acco

I was careful to check the spelling before mentioning their name, unlike you, even when I referred to them earlier. The fact that you cannot even get the position of an underscore in a name, correct, seems to indicate that you are sloppy. which leads me to my next point.

2. pompous:

You said:

>Really, I'm being generous in assuming they didn't reply before you.

This is the pompous bit. Generous? Laughable.

I neither need not want your generosity. If anything, I prefer objectivity, and that people give others the benefit of the doubt, instead of assuming bad intentions by them: I had actually checked for a second comment by _acco (in reply to layer8) just before I wrote my comment to layer8, the one that got all of you in a tizzy. But you not only got the times wrong (see your edit, and point 3 below), but also assumed bad faith on my part.

3. fake.

You first said above that both those replies to layer8 showed as 13 hours ago, then edited your comment to say 14 and 13 hours. It shows that you don't use your brains. The feature of software showing time deltas in the form of "hours ago" or "days ago", versus an exact time stamp, is pretty old by now. It dates back to Web 2.0 or earlier, maybe it was started by some Web 2 startups or by Google.

If you think you are so clever as to criticize me without proof, or say that you are generous in your assumptions about me, you should have been equally clever or generous about the time delta point above, and so realized that I could have replied to layer8 before _acco, which was indeed the case. Obviously I cannot prove it, but the fact that I got _acco's name correct, while you did not, lends credence to my statement. It shows that I took care while writing my comment.

4. So you are fake because you don't bother to think before bad-mouthing others, and even more fake because you did not apply (to yourself) your own made-up "rule" in this other comment of yours, where you criticized my comment as being neither smart nor nice, so not of value:

https://news.ycombinator.com/item?id=41310460

I should not have had to write such a long comment to refute your silly and false allegations, and I will not always do that, but decided to do it this time, to make the point.

And, wow: you managed to pack 3 blunders (being inaccurate, pompous and fake) into a comment of just a few lines. That's neither smart not nice. Instead, it's toxic.

Actually, your inaccurateness (inaccuracy? GIYF) is even worse than I said above. My comment of a few levels above, literally uses the name _acco at least four times - I checked just now. And your comment was in reply to that. So even after reading that person's name four times in my comment, you still got its spelling wrong. Congrats. (Yeah, I can snark too, like you did to me upthread.)

You seem to have gotten so worked up over my misplaced underscore that you yourself forgot how those ubiquitous rounded timestamps work. When I first wrote my comments, they were indeed the same, 3 and later 13 hours. After I wrote my later comment, in the few minutes between times I looked at it, the timestamp on yours just happened to cross the threshold where it rounded up to 14 instead of down to 13. (And if I was "sloppy", do you really think I would have looked again and corrected my comment?) Presumably if I looked at them a bit later they would have both said 14 hours. Hence <1 hour lead.

Anyway, yeah, I worry less about being nice to people who've already shown themselves to be clowns, in a sub thread that's flagged to death. You got me there. FWIW I was originally hoping to enlighten you a bit as to why you were being downvoted, as a small help to you.

  Comments should get more thoughtful and substantive,
  not less, as a topic gets more divisive.

  Eschew flamebait. Avoid generic tangents. Omit internet
  tropes.

  Be kind. Don't be snarky. Converse curiously;
  don't cross-examine. Edit out swipes.

  Please don't comment about the voting on comments.
  It never does any good, and it makes boring reading.

source: https://news.ycombinator.com/newsguidelines.html

HTH

Context is the most challenging bit. FWIW, the codebases I'm working on are still small enough to where I rarely need to include more than 12 files into context. And I find as I make the context bigger beyond that, results degrade significantly.

So I don't know how this would go in a much larger codebase.

What floored him was simply how much of my programming I was doing with an LLM / how little I write line-by-line (vs edit line-by-line).

If you're really curious, I recorded some work for a friend. The first video has terrible audio, unfortunately. This second one I think gives a very realistic demonstration – you'll see the model struggle a bit at the beginning:

https://www.loom.com/share/20d967be827141578c64074735eb84a8

I know that you are getting some push-back because of your exuberance regarding your use of LLMs in development, but let me just say I respect that when someone told you to "put up or shut up" you did. Good on you!

So you spend 10 minutes writing a free text description of the test you want; tell it exactly how you want it to write the test, and then 4-5 minutes trying to understand if it did the right thing or not, restart because it did something crazy then a few minutes manually fixing the diff it generated?

MMmm.

I mean, don't get me wrong; this is impressive stuff; but it needs to be an order of magnitude less 'screwing around trying to fix the random crap' for this to be 'wow, amazing!' rather than a technical demonstration.

You could have done this more quickly without using AI.

I have no doubt this is transformative technology, but people using it are choosing to use it; it's not actually better than not using it at this point, as far as I can tell.

It's slower and more error prone.

Stoked you watched, thanks. (Sorry the example isn't the greatest/lacks context. The first video was better, but the mic gain was too high.)

You summed up the workflow accurately. Except, I read your first paragraph in a positive light, while I imagine you meant it to be negative.

Note the feedback loop you described is the same one as me delegating requirements to someone else (i.e. s/LLM/jr eng). And then reading/editing their PR. Except the feedback loop is, obviously, much tighter.

I've written a lot of tests, I think this would have taken 3-4x longer to do by hand. Surely an hour?

But even if all things were roughly equal, I like being in the navigator seat vs the driver seat. Editor vs writer. It helps me keep the big picture in mind, focused on requirements and architecture, not line-wise implementation details.

> I've written a lot of tests, I think this would have taken 3-4x longer to do by hand. Surely an hour?

I guess my point is I'm skeptical.

I don't believe what you had the end would have taken you that long to do by hand. I don't believe it would have taken an hour. It certainly would not have taken me or anyone on my team that long.

I feel like you're projecting that, if you scale this process, so say, having 5 LLMs running in parallel, then what you would get is you spending maybe 20% more time reviewing 5x PRs instead of 1x PR, but getting 5x as much stuff done in the end.

Which may be true.

...but, and this is really my point: It's not true, in this example. It's not true in any examples I've seen.

It feels like it might be true in the near-moderate future, but there are a lot of underlying assumptions that is based on:

- LLMs get faster (probably)

- LLMs get more accurate and less prone to errors (???)

- LLMs get more context size without going crazy (???)

- The marginal cost of doing N x code reviews is < the cost of just writing code N times (???)

These are assumptions that... well, who knows? Maybe? ...but right now? Like, today?

The problem is: If it was actually making people more productive then we would see evidence of it. Like, actual concrete examples of people having 10 LLMs building systems for them.

...but what we do see, is people doing things like this, which seem like (to me at least), either worse or on-par with just doing the same work by hand.

A different workflow, certainly; but not obviously better.

LLMs appear to have an immediate right now disruptive impact on particular domains, like, say, learning, where its extremely clear that having a wise coding assistant to help you gain simple cross domain knowledge is highly impactful (look at stack overflow); but despite all the hand waving and all the people talking about it, the actual concrete evidence of a 'Devin' that actually builds software or even meaningfully improves programmer productivity (not 'is a tool that gives some marginal benefit to existing autocomplete'; actually improves productivity) is ...

...simply absent.

I find that problematic, and it makes me skeptical of grand claims.

Grand claims require concrete tangible evidence.

I've no doubt that you've got a workflow that works for you, and thanks for sharing it. :) ...I just don't think its really compelling, currently, to work that way for most people; I don't think you can reasonably argue it's more productive, or more effective, based on what I've actually seen.

> I've written a lot of tests, I think this would have taken 3-4x longer to do by hand. Surely an hour?

It seems to me that the first few are almost complete copy-paste of older tests. You would have got code closer to the final test in the update case with simple copy-paste than what was provided.

The real value is only in the filtered test to choose randomly (btw, I have no idea why that’s beneficial here), and the one which checks that both consumers got the same info. They can be done in a few minutes with the help of the already made insert test, and the original version of the filtered test.

I’m happy that more people can code with this, and it’s great that it makes your coding faster. It makes coding more accessible. However, there are a lot of people who can do this faster without AI, so it’s definitely not for everybody yet.

I used it for some troubleshooting in my job. Linux Sys Admin. I like how I can just ask it a question and explain the situation and it goes thru everything with me, like a Person.

How worried are you that it's giving you bad advice?

There are plenty of, "Just disable certificate checking" type answers on Stack Overflow, but there are also a lot of comments calling them out. How do you fact check the AI? Is it just a shortcut to finding better documentation?

In my opinion it’s better at filtering down my convoluted explanation into some troubleshooting steps I can take, to investigate. It’s kind of like an evolved Google algorithm, boiling down the internet’s knowledge. And I’ve had it give me step by step instructions on “esoteric” things like dwm config file examples, plugins for displaying pictures in terminal, what files to edit and where…it’s kind of efficient I think. Better than browsing ads. Lol.

> Certainly, yes. To initialize the IPv6 subsystem, enter 'init 6' in an elevated system terminal.

Whoops.

I think that Greptile is on the right track. I made a repo containing the c# source code for the godot game engine, and it's "how to do X", where X is some obscure technical feature (like how to create a collision query using the godot internal physics api) is much better than all the other ai solutions which use general training data.

However there are some very frustrating limitations to greptle, so severe that I basically only use it to ask implementation questions on existing codebases, not for anything like general R&D: 1) answers are limited to about 150 lines. 2) it doesn't re-analyze a repo after you link it in a conversation (you need to start a new conversation, and re-link the repo, then wait 20+ min for it to parse your code) 3) it is very slow (maybe 30 seconds to answer a question) 4) there's no prompt engineering

I think it's a bit strange that no other ai solution lets you ask questions about existing codebases. I hope that will be more widespread soon.

I work at Greptile and agree on all three criticisms. 1) is a bug we haven't been able to fix, 2) has to do with the high cost of re-indexing, we will likely start auto-updating the index when LLM costs come down a little, and 3) has to do with LLM speed. We pushed some changes to cut time-to-first-token by about half but long way to go.

Re: prompt engineering, we have a prompt guide if that helps, was that what you are getting at?

https://docs.greptile.com/prompt-guide

No idea about the product, but I would like to congratulate you guys on what is maybe the greatest name ever. Something about it seems to combine "fierce" with "cute", so I think you should consider changing your logo to something that looks like reptar

So glad you like it - we were worried it was too silly.

The logo is supposed to be a reptile claw but we might modify it to make that more obvious.

> What exactly floored your colleague at Microsoft?

Speaking of understand context… They floored him not other way round

I think my general perception is that AI is a great assistant for some occupations like software engineering, but due to its large room for error it's very impractical for majority of business applications that require accuracy. I'm seeing this trend at my company, which operates in the medical field and recently mandated that all engineers use CoPilot. At the same time it's a struggle to see where we can improve our business processes with AI - outside of basic things like transcriptions and spell checking - without getting ourselves into a massive lawsuit.

It's useful for analyzing data, but only if it can be verified. We use it (higher education) to glance at data trends that may need further exploration. So it's a fancy pivot table, I guess.

Most software vendors are selling their version of AI as hallucination free though. So that's terrifying.

>Most software vendors are selling their version of AI as hallucination free though.

It definitely has that Tesla Autopolit feel to it, where the marketing is ... enthusiastic, while the onus to "use the tool responsibly" is left up to the end user.

I'm building a persentation/teaching tool for teachers and trying to not sweep the halicination angle under the rug. My marketing angle is that the AI is a huge time saver, but the classroom teacher is the subject expert and they need to review the output. At the end of the day, it's still much more time efficient for an expert to review the content than it is to produce it from scratch which is a huge win.

Sadly, I think your sales pitch will fall flat. College and University admin aren't technology experts, but are VERY susceptible to flattery from tech sales people. They want to hear about the time and cost savings, regardless of reality.

Also, to be super presumptuous, if you need help with anything with your tool, hit me up. I love working with developers on building realistic use case and workflows.

>Sadly, I think your sales pitch will fall flat. College and University admin aren't technology experts

I'm building tools for K-12 (I've been a teacher for 20+ years). There's solid uptake of GPT among teachers and individually they're pretty enthusiastic about anything that will reduce their workload, so I'm hoping it resonates.

>Also, to be super presumptuous, if you need help with anything with your tool, hit me up. I love working with developers on building realistic use case and workflows.

Sure, my details are in my profile, happy to chat.

>I think my general perception is that AI is a great assistant for some occupations like software engineering

I think that's why I like to compare the current state of AI to the state of the CPU industry maybe around the 286-486 era going towards the Pentium.

How does that mandate work? Do they just check if it’s turned on? How many suggestions get accepted?

It's just a requirement that everyone has a license and did 1-hour introductory training. Whether you actually use it or not is up to you, but it's encouraged.

Software engineering doesn't require accuracy?

I was implying that AI can be useful in software engineering because hallucinations are expected and actively addressed by engineers. It's less of an issue in comparison to automating strategic business decisions by what is essentially a "black box" and not knowing if it hallucinated data in order to arrive at its decision.

I've been using (GitHub) Copilot and ChatGPT since they've been widely available. I started using ChatGPT for coding after 4 came out.

I was an early advocate for Copilot but, honestly, nowadays I really don't find it that useful, compared to GPT-4o via ChatGPT.

ChatGPT not being directly integrated into my editor turns out to be an advantage. The problem with Copilot is it gets in the way. It's too easy to unintentionally insert a line or block completion that isn't what you want, or is out and out garbage, and it's constantly shoving up suggestions as I type, which can be distracting. It's particularly irritating when I'm trying to read or understand a piece of code, or maybe do a refactor, and I leave my caret in one position for half a second too long, and suddenly it's ghost-inserted a block of code as a suggestion that's moved half of what I'm reading down the screen and now I have to find my place again.

Whereas, with ChatGPT being separate, it operates at a much less intrusive cadence, and only responds when I ask it too, which turns out to be much more useful and productive.

I'm seriously considering binning of my Copilot subscription as a result.

Copilot was surprisingly one of the worst AI products I’ve used. It keeps suggesting nonsense and the insertions it does in VSCode often mess up very basic things like it starts writing things in the wrong place (or adds a function inside another function) and the result doesn’t even compile. It also seems to have zero understanding of the codebase apart from maybe a few lines before and after. Incredible that this is being sold as a real product at this point

I canned my Copilot sub.

Even where microsoft includes it for free, like their automation tools, its always been jank. I found myself going to GPT4 for better answers which is bad when you spend any time thinking about it.

Have you tried using the shortcut for turning copilot on/off? I know what you mean and in those cases I just turn it off for a second and type freely, then turn it back on.

I see so many people hyping Claude Sonnet + Cursor on Twitter/X, yet in real world usage, I find it no better than GitHub Copilot (presumably GPT 4o) + VScode.

Cursor offers some super marginal UX improvements over the latter (being that it’s a fork of VScode), since it allows you to switch models. But Claude and GPT have been interchangeable at least for my workflows, so I’m not sure the hype is really deserved.

I can only imagine the excitement comes from the fact that cursor has a full-fat free trial, and maybe most people have never bothered paying for copilot?

Apart from autocomplete, I got zero utility from Copilot, whereas I'm doing most coding via prompts in Cursor.

Maybe GPT4o changed things.

Hmm, I've definitely always used paid copilot models.

Perhaps it's my language of choice (Elixir)? Claude absolutely nails it, rarely gives me code with compilation errors, seems to know and leverage the standard library very well, idiomatic. Not the same with GPTs.

In my experience ChatGPT seems particularly bad at Elixir, presumably because there is a comparative lack of published code and discussion about it.

Likely a lack of recently published code. Claude has been trained more recently.

Is there a definite answer on what languages / paradigms LLMs tend to be better at?

Copilot user here, never tried Cursor yet, competition is always good.

did a quick check, it's $20/month, and it has a vim plugin: https://github.com/pasky/claude.vim

going to give it a spin

Copilot plugin is maintained by tpope which is not nothing

I have copilot too; it's good for edits, but one shot complete function generating, I don't find it very good at (while sonnet is imho).

Software engineer for multiple decades here. None of the AI assistants have made any major change to my job. They are useful tools when it's time to write code like many useful tools before them. But the hard work of being a senior+ software engineer comes before you start typing.

I will admit, if I need to do some one off task and write a quick python script to do something I will likely go to Claude or something and write it. I am talking 20-40 lines. I think it's fine for that, it doesn't need a ton of context, it's easy to test, easy to look at and understand, etc.

But outside of that, beyond needing to remember a certain syntax, I have found that any time I tried to use it for anything more complex I am finding myself spending more time going back and forth trying to get code that works than I would have if I had just done it myself in the first place.

If the code works, it just isn't maintainable code if you ask it to do too much. It will just remove entire functionality.

I have seen a situation of someone submitting a PR, very clearly copying a method and sticking it in AI and saying "improve this". It made changes for no good reason and when you ask the person that submitted the PR why they made the change we of course got no answer. (these were not just Linter changes)

Thats concerning, pushing code up that you can't even explain why you did something?

Like you said with the hard work, sure it can churn out code. But you need to have a complete clear picture of what that code needs to look like before you start generating or you will not like the end result.

They help me most when using a framework, API, or language I'm not super familiar with. Beats stack overflow for that stuff.

But it's weird to me seeing people talk about these changing their jobs so much. Maybe I'm holding it wrong but I'm almost always bottlenecked by "big picture" challenges and less on the act of actually typing the code.

My GH copilot kept recommending incorrect things when I use it along side common libraries and frameworks, I don't know, I just don't really find it very useful.

That's because programming isn't just about syntax. The real difficulty lies in the problem solving. Sure AI can help with that, but the problem still largely rests on the programmer, as AI won't do anything unless prompted to.

I am not a software engineer and these tools allow me to make a giant mess of an app in a weekend that kind of does what I want but I only get that weekend. Once I come back to it after any length of time, since I have no idea what I am doing or what is going on it is impossible to update or add to the app without breaking it.

Now I have all these new ideas but I am back to square one because it just seems easier to start over.

I look forward to more powerful models in the future but I do wonder if it will just mean I can get slightly farther and make an even larger mess of an app in a weekend that I have no way to add to or update without breaking.

The main utility seems like it would be for content creation to pretend I made an app with all these great features as a non-software engineer but conveniently leave out the part about it being impossible to update.

This.

It’s great at helping me automate things I normally do not have time to attempt.

I think the non obvious benefit is using LLMs nudge you into putting your thoughts in narrative form and training that ability, something that someone with more experience does subconsciously.

What do you see as your biggest wins using Claude?

It helps me stay in flow by keeping me one layer up.

In pair programming, it's ideal to have a driver (hands on keyboard) and a navigator (planning, direction).

Claude can act as the driver most of the time so I can stay at the navigator level.

This is so helpful, as it's easy as programmers to get sucked into implementation details or low-level minutiae that's just not important.

> yet their company is barely even leveraging it

...do you not see the obnoxious CoPilot(TM) buttons and ads everywhere? It's even infected the Azure Portal - and every time I use it to answer a genuine question I have I get factually-incorrect responses (granted, I don't ask it trivial or introductory-level questions...).

I said leveraging, not hawking!

How did it "change your job description"?

As I mentioned in a sibling comment, I now "pair program" all day. Instead of being the driver and navigator all day, I can mostly sit "one layer up" in the navigator seat.

How is this a change in job description? It may be a failure of imagination on my part, but it sounds like you're still doing what you have always done - you've changed the how.

Right. And in a similar vein, its very easy to replace a lot of low-effort outsourcing / junior-dev assignable tasks. I still need to describe the problem and clean up the answer, but I get the answer immediately and I can re-prompt for it to fix it immediately.

Do you feel yourself becoming disconnected from your skills? I used gpt extensively at one stage and I felt very rusty getting my hands dirty in code again.

I don't. But that's probably because I'm very opinionated about the implementation details, so I'm scrutinizing and tweaking its output a lot.

What is different vs. GPT-4o & Copilot?

Coding assistance ain’t useful till it has around 50mb of context or whatever the size of 80% of code bases uses

I feel like I'm living in a different universe sometimes. The consensus on HN seems to be that you can be pretty productive with LLMs as coding assistants, but every time I try I find it borderline impossible to get functional code even for pretty straightforward prompts.

I decided to fire up GPT-4o again today to see if maybe things have gotten better over the past few months.

I asked GPT to write code to render a triangle using Vulkan (a 3D graphics API). There are about 1000 tutorials on this that are almost certainly in GPT-4's training data. I gave GPT two small twists so it's not a simple case of copy/paste: I asked it 1) to apply a texture to the triangle and 2) to keep all the code in a single function. (Most tutorials break the code up into about a dozen functions, but each of these functions are called only once, so it should be trivial to inline them.)

Within the first ten lines, the code is already completely nonfunctional:

GPT-4o declares a pointer (VkPhysicalDevice) that is uninitialized. It queries the number of graphics devices on the host machine. A human being would allocate a buffer with that number of elements and store the reference in the pointer. GPT-4o just ignores the result. Completely ignores it. So the function call was just for fun, I guess? It then tries to copy an entire array of VkPhysicalDevice_T objects into this uninitialized pointer. So that's a guaranteed memory access violation right off the bat.

I feel the exact same way. I have felt 0 need to use an LLM in my current workflow. If I could explain concisely what my problem is in English - then I would already know the answer. In that case why would I be asking AI the question. And I don't want opinionated answers to subjective questions. Why run that through a filter, I can investigate better myself through links on the internet.

By the way - I think AI or ML whatever has some valid uses right now. but mostly in image processing domain - so like recognizing shapes in some bounded domain OK yea. Generative image - NOT bad but theres always this "AI GLOW" to each image. Something is always off. Its some neat tools but a race to the bottom and mostly users want to generate explicit content lets be real. and they will become increasingly more creative and obtuse to get around the guards. nothing is stopping you from entering the * industry and making tons off money. but that industry is always doing good.

a friend recently suggested to use AI to generate generic icons for my game. Thats a really good use case. but does that radically change the current economy?

>> If I could explain concisely what my problem is in English - then I would already know the answer. In that case why would I be asking AI the question. And I don't want opinionated answers to subjective questions. Why run that through a filter, I can investigate better myself through links on the internet.

I am an experienced programmer, but I find myself (for the first time) doing a deep-dive in SQL and specifically building code that will run against multiple SQL engines.

I fed my list of engines into AI. As I'm discovering the really weird corners of SQL I asking the AI to compare one db against the other. My prompts are usually no more than 3 or 4 words.

It gives me quick helpful answers highlighting where things are the same and where they are different. I can then follow up in the specific docs if necessary (now that I know the function name.)

Personally I'm somewhat anti-hype, I'll let others rave about "changing the world". But I have found it a useful tool - not so much for "writing my code" but for acting as my tutor as I learn new things. (I'm using it for more than just SQL or computers now.)

I'm not sure it "changes thr economy" - but it can certainly change individual lives. Some jobs will go away. Others might be easier to do. It might make it easier to learn new skills.

Recently I saw the process of filling for 2 insurance claims through 2 different entities. First one used a great ai voice agent that does the process of filtering your query. it understood me perfectly. But then I still had to wait in line for an actual agent. Ok wait for 6 hours. Whatever. Call again - the 2nd time going through the same agent is painful. Its so slow. And all this just to connect me to a human - who was EXCELLENT. What did ai add? And for now a human has to be involved in any complex permission changes to your account. AND I like that.

The 2nd agency I called them through phone. No ai. But it is excellent cause they do async processing. so you reserve a slot and they call you back. i don't care if those are not answered urgently. Because I just want to talk to a human.

AI lets the agency better be able to afford to hire and keep the excellent humans around to solve complex problems. It's better than a human that also can't fix your issue themselves and has to ask you to transfer to the other person. Ideally, the AI agent would have let you book a callback time with the specific right human who's best able to fix your issue. Some companies are able to do this, and you, the customer, never realize that of the 100 specialists in the building, you got connected to the one person who understands your issue and has the tools to fix it.

Customer service is hard because it has to filter out 90% noise, 9% fairly straightforward tasks, and the <1% of complex issues that need to be sent to the right human.

DONT get me wrong. What you're saying I actually support. If this allows that company to have quality staff I'm all for it. The less time we both spend on each other the better. So just offer async processing.

Your point is that AI is bad at some things, and in some cases misused. Which is of course abundantly true.

But equally it doesn't prove, or even assert, the opposite. A bicycle may be bad at cross-country road trips, but that doesn't make it a bad choice for some other situations.

Hence my earlier comment - I (and I suspect others) are finding it useful for some tasks. That is not to imply it is good at all tasks.

Is it over-hyped? Of course yes. Welcome to IT where every new thing is over-hyped all the time.

I think llms are useful when you're trying to write something in a language you don't know well; then it speeds up the part where you need to check for simple idiosyncrasies.

If you don't know the language at all it's dangerous because you may not understand the proposed program (and of course if you're an expert you don't need it at all).

But llms won't help to find solutions to a general, still unspecified problem.

LLM coding assistants are more like another iteration in the jump from nothing -> Stack Overflow type resources than a replacement for you doing coding work as a programmer just because you can phrase the task as a prompt. If you measured the value of Stack Overflow posts by blindly merging the first 2 hundred line examples of the things you want to do, finding the result wasn't ideal, and declaring it a useless way to get functional code people would (rightfully) scratch their heads at you when you say they are living in a different world.

Essentially to me it feels like almost all LLM's are either mid -> terrible if it's any system programming. Especially I've not had luck with anything outside of web programming and moreso anything NOT javascript / python. (jeez I wrote this sentence terribly)

For LOVR (Lua based 3D/VR framework), I found these LLMs pretty much useless, both ChatGPT and Claude. Seems all is trained on old APIs, so it takes quite a bit of effort to make any suggestions work with newest LOVR version.

I wonder if quality would improve if one would upload the latest LOVR documentation and uploaded LOVR codebases using the newest version and instruct it properly?

The docs are available online. I don't know how often models are "re-trained".

Chat gpt has helped me with some complex sql queries. I had to correct it a couple of times but in the end it worked.

Chat GPT seems really good at frequently asked questions.

Like simple python scripts for Home Assistant it just nails first go.

It has helped, at the cost of making me more dependent on the assistants

I'm guessing AI won't be great with anything nontrivial related to pointers for a while since they require actual thinking

At least, we can hope that AI will master pointers one day. For people it's completely hopeless

Give it another shot but with Claude Sonnet 3.5. It’s my daily driver for coding tasks.

It seems especially strong with Python but a bit medium with Swift.

I just signed up for the free version. Claude Sonnet does properly use malloc/free to manage the buffer where GPT-4o screws up (Yay!) It manages to gets through the whole process of initializing a graphics device and grabbing a queue from the device. It took some questionable shortcuts to get there (and didn't leave any comments explaining those shortcuts and the problems they could cause down the road), but fine, the code works.

After that it goes completely off the rails by trying to issue draw commands before binding a graphics pipeline, which is both illogical and illegal. After a lot of prodding, I did manage to get it to bind a graphics pipeline, but it forgot about the texture.

So Claude Sonnet is definitely better than GPT-4o, but it still feels raw, like a game of whack-a-mole where I can get it to fix a mistake, but it reintroduces an old one. I also have to be the one offering the expertise. I can prompt it to fix the issues because I know exactly what the issues are. If I was using this to try to fill in for a gap in my knowledge, I would be stuck when I ran the code and it crashed - I would have no idea where to go next.

Update: Took about 50 min of experimenting, but I did get Claude to generate code that doesn't have any obvious defects, although it cut off about halfway through because of the generation limit. The best result that I've seen from an LLM yet. But that's after about a dozen broken programs, and again, I think the domain expertise here is key to be able to reprompt and correct.

I've had some moderate success asking Claude to "translate" smallish pieces of code from, eg, C++ to Python. One simple C++ file parser it managed to translate basically 100% right on one try. I wouldn't particularly trust the output- well, not until I run it through tests- but for quick "could this possibly work, what does the performance look like, what might this code look like" exploratory stuff it's been very useful, especially for code that you are likely to throw away anyway.

One example that I am still using: I wanted to generate a random DICOM file with specific types of garbage in it to use as input for some unit tests, and Claude was able to generate some Python that grabs some random DICOM tags and shoves vaguely plausible garbage data into them, such that it is a valid but nonsensical DICOM dataset. This is not hard, but it's a lot faster to ask Claude to do it.

LLMs are much better at Python and JavaScript than at C/C++. This simple difference can account for much of the variation in people's experiences.

I agree, that could explain a lot of it. I also suspect that the length of the generated code plays a role. In my experience, LLMs sometimes peter out a bit and give up if the generated program gets too long, even if it's well within their context limit. Python and JavaScript tend to be more succinct and so that issue probably comes up less.

Yes, you have figured it out. LLMs are terrible for graphics programming. Web development - much better. Sonnet 3.5 is the only good model around for now. GPT 4o is very poor.

Yeah, these people claiming AI has been a transformative experience are just full of sh*t. I ask various models questions all the time because it's often better than googling, but all of them make a lot silly mistakes. Frequently, it can be a bit of a process to get useful results.

Its not supposed to do your work. Its an assistant.

... Or they just honestly have a different experience than you

Try deepseek coder

My experience with CoPilot has been similar. I don't think it's given me a single piece of code that just worked. It always took several back and forth of me telling the "AI" that I need it to do X instead of Y (which was included in the original instructions but ignored).

It seems to work best if I start with something very simple, and then layer on instructions ("now make it do X").

Where I have found it saves me time is in having to look up syntax or "gotchas" which I would otherwise search StackOverflow for. But as far as "writing code" -- it still feels a long way from that.

I've almost gotten GPT-4o to do some accurate regex. Probably 75% of the time it's close enough that I can tweak it and make it work.

I feel the same way. Anytime someone says they don't find LLMs all that useful, the exact same comments come out:

"They clearly aren't using the right model!"

"It's obvious they don't know how to prompt, or they would see the value."

"Maybe it can't do that today, but GPT-5 is just around the corner."

I feel more and more that people have just decided that this is a technology that will do everything you can imagine, and no evidence to the contrary will change their priors.

For me llm is faster stackoverflow without fear of my question being closed. I know the first answer will not be what I want. I know that I will have to refractor it to suit my style. I know it will be full of subtle bugs.

Oh and I expect it to be free, I ain't paying for this just like I wasn't paying for stackoverflow.

Finally I hope than in few years I will be able to just "sudo apt-get install llm llm-javascript llm-cooking llm-trivia llm-jokes" and it will all run locally on my low end computer and when I report bug, six months later it will be fixed when I update OS.

You are paying, the same way you paid for stack overflow.. you become part of the process, ask follow up questions, deepening the knowledge of the system.

The same applies to AI. The old learning material is gone, your interaction is now the new learning material and ground truth.

PS: Hourly rates for sw engineers: Range:€11 - €213 - so one hour on stackoverflow, searching and sub-querying resolving problems costs you or your employer up to 213€. It really depends on what you have negotiated.

And unlike Stack Overflow, which is available to everyone online and has an open content license, the IP the users of the ChatGPT style services is entirely proprietary to the company. I am not interested in feeding their machine with my brainpower. On the other hand, I happily contribute to Stack Overflow, and open source software+hardware. I do not think I will integrate LLM into my engineering workflow until there is a good service/solution which builds up the commons. The huge companies already have way too much influence over key aspects of knowledge-based society.

> I am not interested in feeding their machine with my brainpower

How about DDG AI Chat? https://duckduckgo.com/?q=DuckDuckGo+AI+Chat&ia=chat&duckai=...

Turn it off in the settings, then... Come on.

Every month, every update, every week, every contract change, every day - the setting is in another menu, in another castle, the setting is a princess, kiss the button, the button is now on, now off, is now dark patterned, now it isn't, deliver proof of work to set the setting, solve 1 captcha, solve 20.. come on.. its rodeo time.

You know you can just turn off "Improve the model for everyone" in the settings, right?

I'm with you. Every time I've used LLMs in my work, I've ended up using more time tidying up after the robot than it would have taken to just do the work myself from the start. I can believe that there are some tasks that it can do very fast, but my experience is that if you're using it for anything that matters, you can't trust it enough to let it do it on its own, and so you just end up doing the work twice.

It's like having an unusually fast-but-clumsy intern, except interns learn the ropes fast and understand context.

More likely to be confirmation bias; most of us ask the wrong questions, try to confirm what we already believe rather than choose questions that may falsify our beliefs.

I have some stand tests for LLMs: write a web app version of tetris, write a fluid dynamics simulation, etc., and these regularly fail (I must try them again on 4o).

But also, I have examples of them succeeding wildly, writing a web based painting app just from prompting — sure, even with that success it's bad code, but it's still done the thing.

As there are plenty of examples to confirm what we already believe, it's very easy to get stuck, with nay-sayers and enthusiasts equally unaware of the opposite examples.

I mean when the people saying it doesn't work for them, would kill them to give links to the chats on ChatGPT.com so everyone can see the prompts used? when they do, it's a different conversation. like the number of R's in strawberry, or 9.11 - 9.2. When the complaints are generic with no links, the responses are similarly generic, because both sides just biases that they're right and the other side is the one that's wrong.

I welcome people picking apart chats that I link to. it's not that I believe that LLMs are magic and refuse to adjust my model of how good these things are and aren't, but when people don't give specific evidence is hard to actually move the conversation forwards.

because yeah, these things are plenty stupid and have to be tricked into doing things sometimes (which is stupid, but here we are). they're also pretty amazing but like any hammer, not everything is a nail.

It codes great for me, helps me deliver faster, better tested and more features. It literally saves time and money every day. If you can't do that, maybe, just maybe, you have a you problem. But there are many like you in this thread so you are not alone.

The worst part about LLMs is this attitude it's giving to people who get a few helpful answers in a row

You're like a gambling addict who thinks he's smarter than everyone else

Can you state your domain and programming language for clarity?

JS/React and SAP/Hybris at work, vanillaJS for side projects. Ecommerce and general web/app development

Same here.

A friend, whose SQL knowledge is minimal, used an LLM to query data from a database over a couple of tables. Yes, after a lot of trial and error he (most probably) got the correct data, however the only one being able to read the query is the LLM itself. It's full of coalesce, subselects that repeat the same joins again and again.

LLM will do a lot for you, but I really hate this "this will [already did] solve everything". No, it did not and no, because it's quality is those of a junior dev, at max.

> because it's quality is those of a junior dev, at max.

Echo chamber/bias I guess. I know many, many seniors with big pay checks working for big companies who are vastly worse than sonnet at the moment. Juniors stand literally no chance unless very talented.

I understand a junior dev as someone who's missing (or does not have at a greater level) of maintainability, readability, longevity of a solution.

I don't know what companies pay for and I don't care, because if we go by that, every definition of every word is arguable (because there's everywhere someone out of the range of the definition)

I was skeptical like you, but recently decided to try it out. I wasn't expecting much, and as such I was slightly surprised.

For example, just now my NAS stopped working because the boot device went offline. So I got to thinking about writing a simple syslog server. I've never looked at the syslog protocol before, and I've never done any low-level TCP/UDP work in C# yet.

So I asked ChatGPT to generate some code[1], and while the result is not perfect it's certainly better than nothing, and would save me time to get going.

As another example, a friend who's not very technical wanted to make an Arduino circuit to perform some automated experiment. He's dabbled with programing and can modify code, but struggles to get going. Again just for kicks, I asked ChatGPT and it provided a very nice starting point[2].

For exploratory stuff like this, it seems to provide a nice alternative to searching and piecing together the bits. Revolutionary is a quite loaded word, but it's certainly not just a slight improvement on what we had before LLMs and instead feels like a quantum leap.

[1]: https://chatgpt.com/share/f4343939-74f1-404d-bfac-b903525f61... (modified, see reply)

[2]: https://chatgpt.com/share/fc764e73-f01f-4a7c-ab58-f43da3e077...

This is how i currently utilise and view The AI tools. I replaced the googling in my flow with it. It might have taken me 20minutes before to get the boilerplate together, now its a minute. Depending on the type of code legacy overheads you have, results may vary. If you can break your problem down into small descrete bits.

The way the OP here was talking about sonnet being above way above chatgtp in this case, it could be true. Google probably has the largest Go codebases on search to train the AI on higher quiality inputs. Go is a simpler language with less variation over time compared to something like .net also adding to its corner.

ive always been the type of person to use the right language for the each use case. For the last 10+ years ive primarly been building cross platform apps that target every common OS. So these "ai" tools like phind.com give me a universal API interface and generate code which is equiv to a SO answer. They are the ability of an outsourced junior dev who you would never let push code to prod, that doesnt have the language barrier but retains the fake dergee overheads ;)

Very interesting. A couple of notes here on the C# version.

Its using the old format where the Program.cs file has an actual class, whereas as of .NET 6 thats not required.

You said barebones, but for any real server you would want to use the generic host https://learn.microsoft.com/en-us/dotnet/core/extensions/gen... which gets you a lot of the boilerplate and enables you program to be wrapped in a windows or systemd service.

Finally, parsing can be simplified since ASCII is a proper subset of UTF-8, you can just parse the entire string as UTF-8. IMHO I am disappointed that the AI didn't point that out.

> You said barebones, but for any real server you would want to use the generic host

True, I intentionally said barebones as I wanted a minimal example. I asked it to modify the code to use the generic host, and updated the chat link (so refresh). Keep in mind this is the free ChatGPT, but I still think it did reasonably good. The example compiles as-is, and is very close to functional. I've not used the generic host stuff before either, so again this would save me time searching and piecing together code.

> Finally, parsing can be simplified since ASCII is a proper subset of UTF-8, you can just parse the entire string as UTF-8.

I don't think that would work, because the free-form text message part at the end must contain a BOM if it's UTF-8 encoded, according to the specification. AFAIK you can't have the BOM in the middle of a string.

This is a perfect example of the llm producing an answer that a novice finds useful and an expert takes issue with.

This is what the AI is great at - the topics might be obscure to you but they are not that obscure in general, so the AI has had a lot of training data.

I've also assembled a Kubernetes cluster overnight, despite not knowing much about Kubernetes before, and I ran the kubectl files ChatGPT made for me past some devops folks, and it passed the smell test.

I consider much of coding to be a magic spellcasting tutorial - we do conceptually simple things, and the difficult lies in figuring out how to use novel libraries and get them to do what you want.

Edit: After checking out the Arduino sketch, I'd take issue with all the floating point calculations in there - most microcontrollers don't have FPUs, and the performance is awful on 8-bit AVRs. It's not great on Cortex M3s either as all this stuff is done in software, and each FP operation is like a hundred cycles.

I'd definitely try to rephrase the issue with integer math. It might work, but no self-respecting embedded dev would write like this.

> the topics might be obscure to you but they are not the obscure in general

Exactly, it's a very nice alternative to searching the web and discovering new stuff.

> most microcontrollers don't have FPUs, and the performance is awful on 8-bit AVRs.

I used to think like you. But then I realized the Atmega 328p is running at 16 MHz, so even hundreds of . As you can see here[1], it can do 94k double-precision FLOPS, more than enough for simple sketches like this. This jives with the benchmarks I did several years ago.

Sure if I was writing some tight control loop or similar I wouldn't use floating point.

[1]: https://kreier.github.io/benchmark/LinpackDP/

Still it does not sit right with me - it might work, but figuring out the math with fixed point is not that hard, just requires some sitting down and thinking about it.

It's like I have some light fixtures in my attic that are connected with wires twisted together and covered with electrical tape - they certainly work and have done for a decade, but its still not right and I wouldn't recommend anyone do it this way.

I’m similar: LLM can improve my productivity by 10-20% when I’m working on something familiar, 30-50% when I’m venturing into a popular domain that I’m unfamiliar with.

I just don’t understand where the hell is this magical LLM capable of generating flawless files or even entire projects that many people are talking about. I rarely accept a large block of LLM-generated code without close inspection, and I’ve ended up with a subtle bug that I wouldn’t have written myself at least ~5 times now. Unless you don’t give a shit about code quality, error handling, proper logging, and subtle bugs, you shouldn’t run LLM-generated stuff without painstakingly reading and fixing everything. Or maybe there really is a magical LLM somewhere.

> but every time I try I find it borderline impossible to get functional code even for pretty straightforward prompts.

I work in 9 different projects now and I would say that around 80% of functional code comes from Sonnet (like GP) for these projects. These are not (all) trivial either; there is a very niche (for banking) key/value store written in Go for instance which has a lot of edge cases etc, all the plumbing (x,err = etc aka stuff people find annoying) comes from sonnet and works one-shot. A lot of business logic comes from sonnet too; it works but usually needs a little tweaking to make it correct.

Tests are all done by Sonnet. I think 80% is low balling it on Go code really.

We have a lot of complex code generator stuff and DSLs in TS which also works well often. Sometimes it gets some edge cases wrong, but either re-prompting with more details or fixing it ourselves, will do it. At a fraction of the time/money of what a fully human team would deliver.

I wrote a 3d editor for fun with Sonnet in a day.

I have terrible results with gpt/copilot (copilot is good for editing instead of complete files/functions; chatgpt is not good much compared with sonnet); it doesn't get close at all; it simply keeps giving me the same code over and over again when I say it's wrong; it hardcodes things specifically asked to make flexible etc. Not sure why the difference is so massive all of sudden.

Note: I use the sonnet API, not the web interface, but same for gpt so...

I'd like to see that 3D editor. It's a highly non-trivial thing to do properly.

It was a few hours work so it was not 'done properly', but the point was, it is working code and, as I know nothing about 3d or game programming, I didn't do any of it. People here claim it (current LLMs) cannot produce working code for entire software's without a human doing a lot of it; it clearly can and for non trivial stuff. I would say the real point is ; it cannot do complete code for non trivial programs without human doing most of the coding IF the person prompting is not an experienced coder. I am, so I can review the generated code, prompt in English what it should change and then it works. It is often faster to write code instead of that (so the thing blurbs 100% working code but with 20% wrong/bugs which I fix as a human) but in domains I know nothing about, English is faster for me.

Something worth noting is that the parent comment refers to using Cursor, not ChatGPT/Claude.ai. The latter are general-purpose chat (and, in the case of ChatGPT, agentic) applications.

Cursor is a purpose-built IDE for software development. The Cursor team has put a lot of research and sweat into providing the used LLMs (also from OpenAI/Anthropic) with:

- the right parts of your code

- relevant code/dependency documentation

- and, importantly, the right prompts.

to successfully complete coding tasks. It's an apple and oranges situation.

Counterpoint: I have gotten real value out of dumping broken SQL into ChatGPT and have it fix it for me. I could 100% have done that by myself, but it would have meant I have to go and google the right syntax.

AI is great for me, but it is more like a junior developer you are pairing with than a replacement.

I've been very productive using LLMs, without any expectations of them "writing functional code". Instead, I've mostly used them as if I was working with a human research librarian.

For example, I can ask the LLM things like "What are the most common mistakes when using the Vulkan API to render a triangle with a texture?" and I'll very rapidly learn something about working with an API that I don't have deep understanding of, and I might not find a specific tutorial article about.

As another example, if I'm an experienced OpenGL programmer, I can ask directly "what's the Vulkan equivalent of this OpenGL API call?" and get quite good results back, most of the time.

So I'm asking questions where an 80% answer is still very valuable, and it's much faster than searching for documentation and doing a lot of comparison and headscratching, and it works well enough even when there's no specific article I could find in a variety of searches.

Anything better that the technology gets from here just makes things even easier yet!

Lots of companies are directly involved with LLMs, or working to leverage it for startups or their existing projects. And I think a fair chunk of all the employees working at these places probably post on HN (a crazy high percent of the recent batches of YC applicants were using LLM stuff, for instance). That's going to lead to a sizable number of perspectives and opinions that are not especially free of bias, simply because you want to believe that what you're working on is ultimately viable.

And I think it'd be extremely easy to convince oneself of this. Look at where 'AI' was 5 years ago, look at where it is today and then try to imagine where it will be in another 5 years. Of course you have to completely blind yourself to the fact that the acceleration has clearly sharply stalled out, but humans are really good at cognitive dissonance, especially when your perception of your future depends on it.

And there's also the point that even though I'm extremely critical of LLMs in general, they have absolutely 'transformed' my workflow in that natural language search of documentation is really useful. Being able to describe a desired API, but in an overly broad way that a search engine can't really pick up on, but that an LLM [often] can, is just quite handy. On the other hand, this is more a condemnation of search engine tech being frozen 20 years in the past than it is about an imminent LLM revolution.

>acceleration has clearly sharply stalled out

I would rather say that the innovation happens in spikes. I see no slowing down whatsoever.

There’s a lot of people on here with a financial interest in AI doing well. So they hype it continuously.

Yes me too! I don't have any stake for finsing LLMs unhelpful, and would love to have a tool to make me more productive.

Would be really interesting if anyone had blog posts on their actual workflow with LLMs, in case there's something I'm doing different.

one trick is to let it come up with what it wants (lots of functions, no texture), then run the code, give it the errors until that's fixed. then ask it to inline them, then add the texture, etc.

Try something like maestro which uses agents and orchestrator. The looping quality checks have been very helpful. Claude-engineer from same developer is also good to experience how superior that ia to regular chat.

I'd go one step further and suggest that one of the few things LLMs are good at is acting as astroturfing agents for themselves. I must be crazy to not be completely changing my workflow and job title with [THE_GOOD_ONE] and not that other one, and wow so many other people vocally feel the same way on every other forum that's extremely easy to manipulate.

Fwiw, I've had some helpful successful prompts here and there, and in some very narrow scopes I'll get something usable, like parsing JSON or scaffolding some test cases, which is real saved time, but I stopped thinking about these tools long ago.

To get real value out of something like your example, I'd be using it as a back and forth to help me understand how some concepts work or write example questions I can drill on my own, but nothing where precision matters

Thats because GPT products are in a different (much worse) universe to Anthropics Sonnet/Opus, they are truly phenomenal.

Give Anthropic a shot (its even better via the API console.anthropic.com/workbench).

OpenAI is yesterdays news.

Where LLM shine is as a faster alternative to Google and Stack overflow. "How do I reverse an array in language X?". This will give you the right answer in seconds without having to click through garbage.

Especially if it's a question that's hard to Google, like "I remember there is more than one way to split an array in this language, list them". This saves me minutes every day.

But it's especially helpful if you are working on projects outside your own domain where you are a newbie.

> This saves me minutes every day.

That is very believable, the people who say it made them super productive is a bit less believable.

Gpt seems to have gotten worse, Claude is the new hotness.

But, I agree with your sentiment that asking it to do stuff like that often doesn’t work. I’ve found that what it _can_ do is stuff like “here’s a Model object, write a query to fetch it with the schema I told you about ages ago”. It might not give perfect results, but I know how to write that query and it’s faster to edit Claude’s output than it is to write it from scratch.

> I feel like I'm living in a different universe sometimes. The consensus on HN seems to be that you can be pretty productive with LLMs as coding assistants, but every time I try I find it borderline impossible to get functional code even for pretty straightforward prompts.

Same, it can't even fix an xcode memory leak bug in a simple app. It will keep trying and breaking it non-stop. Garbage

Sometime I think there's something wrong with me. I've used copilot, I'm paying for ChatGPT and we're also having the Jetbrains AI, and there's just something so off about all of this.

Some basic things are fine, but once you get into specialised things, everything gets just terribly wrong and weird. I can't even put it into words, I see people saying how they are 10x more productive (I'd like to see actual numbers and proof for this), but I just don't see how. Maybe we're working on very custom stuff, or very specific things, but all of these tools seem to give very deep or confident answers that are just plain wrong and shallow. Just yesterday I used GPT 4o for some basic help with Puppet, and the examples it printed, even though I would say it's quite basic, were just wrong, but in the sense of having to debug it for 2 hours just to figure out how ridiculous the error was.

I fear the fact that people will end up releasing unsafe, insecure and simply wrong code every day, code that they never debug and not even understand, that maybe works for a basic set of details, but once the real world hits it, it will fail like those self driving cars driving full speed into a trailer that has the same color as the road or sky.

It's going to be the same issue as poorly-structured Excel workflows: non-programmers doing programming. Excel itself gets the blame for the buggy poorly-structured spreadsheets made by people who don't have the "programmer mindset" of thinking of the edge cases, exceptions, and extensibility.

So it will be with inexperienced people coding with LLMs.

Absolutely the same for me. Whenever I encounter a problem I'm pretty sure nobody else has encountered before (or has not written about it at least), ChatGPT writes complete nonsense.

On stuff that's a bit obscure, but not really (like your Vulkan example), ChatGPT tends to write 60-95% correct code, that's flawed in the exact ways a noob wouldn't be able to fix.

In this case, nicking code from Github seems to fix the issue, even if I need to adapt it a bit.

Then comes the licensing issue. Often, when searching for an obscure topic, the code ChatGPT generates is very close to what's found on Github, but said code often comes with a non-permissive license, unlike what the AI generates.

> Then comes the licensing issue. Often, when searching for an obscure topic, the code ChatGPT generates is very close to what's found on Github, but said code often comes with a non-permissive license, unlike what the AI generates.

I think this is a feature for a lot of people. ChatGPT can launder code so they don't have to care about licenses.

Having spent a couple of hours with some llama models and others I've given up on them for code. Code generation from XML or grinding it out ad hoc with stuff like sed on top of a data or config file is faster and more convenient for me.

The Jetbrains thing is rather rudely incompetent, it consistently insists on suggestions that use variables and fragments that are supposed to be replaced by what I'm writing and also much more complex than what I actually need. I suffered through at least a hundred mistaken tabbed out shitty suggestions before I disconnected it.

I think there's a vast ocean of different software engineers and the type of code they write on a daily basis, which is why you get such differing views on AI's effectiveness. For me, AI has only ever been useful for very basic tasks and scripts: If I need a quick helper script for something in Python, a language that isn't my daily drier, AI usually gets me there in a couple of prompts. Or maybe I am writing some powershell/bash and forgot the syntax for something and AI is quicker (or has more context) than a web search.

However, my main job is trying to come up with "elegant" architectures for complex business logic that interacts with an existing large code base. AI is just completely out of its depth in such cases due to lack of context but also lack of source material to draw from. Even unit tests only work with the most basic of cases, most of the time the setup is so complex it just produces garbage.

I've also had very little luck getting it to write performant code. I almost have to feed it the techniques or algorithms before it attempts to write such code, and even then it's usually wrong or not as efficient as it could be.

Indeed, if you need a lot of boilerplate that's pretty similar to existing commonly available code, you're set. However...

That code is probably buggy, slow, poorly architected, very verbose, and has logical issues where the examples and your needs dont exact match.

Generally, the longer the snippet you want your LLM to generate, the more likely its going to go off the rails.

I think for some positions this can get you 90% of the code done. For me this usually means I can get started very fast on a new problem, but the last remaining "10%" actually takes significantly longer and more effort to integrate because I dont understand the other 90% off the top :)

> my main job is trying to come up with "elegant" architectures for complex business logic that interacts with an existing large code base

Right, that's surely the main job of nearly every experienced developer. It's really cool that LLMs can generate code for isolated tasks, but they can barely even begin to do the hard work, and that seems very unlikely to change in the foreseeable future.

I have the same experience. I have my own benchmark, I take a relatively complex project like FreeSWITCH on github, which is part of the training set for all coding LLMs anyway so they should know it and I ask the AI to code small snippets, tests, suggestions and fixes to see how well it understands the codebase and the architecture.

I just tried the latest Cursor + Sonnet and it failed in every task. The problem is that there is no way to understand the code without either complete understanding the domain and the intents or running it in some context.

Telecom and media domains in particular are well documented in specs and studied in forum discussions. I am sure they are part of the training data, because I can get most answers if asked directly. So far the LLMs fail to reason about anything useful for me.

I don’t use ChatGPT or chat AI for coding, at least not often. It does help me get unstuck on occasion, but most of the time the context switch is too much of a productivity sink.

However, I use copilot autocomplete consistently. That makes me much more productive. It’s just really good autocomplete.

As an experienced developer I’d put my productivity improvement at 2-3x. It’s huge but not 10x. I’m limited by my decision speed, I need to decide what I want the code to do, AI can’t help with that - it can only do the “how”.

Far from introducing more bugs, using Copilot frees some mental cycles for me to be more aware of the code I’m writing.

I had a decent experience putting together frontend code for a demo control system with Mixtral within a couple days. I'm a seasoned programmer but I don't do JS. It stumbled a dozen times but it fulfilled the task of me avoiding to learn JS.

However once you step outside JS or Python, the models are essentially useless. Comprehension of pointer semantics? You wish. Anything with Lisp outside its training corpus of homework assignments? LOL. Editing burden quickly exceeds any possible speed-up.

I look at this contrast as what I call the difference between a "programmer" and a "software engineer". These jobs are really two different universes in practice, so you're not wrong.

I saw an LLM demo at one point where it was asked to write FFT and add unit tests for it which really drove this point home for me.

A programmer is a nicer term for a code monkey. You ask them to write FFT and they'll code it. All problems can be solved with mode code. They can edit code, but on the whole it's more just to add more code. LLMs are actually pretty good at this job, in my experience. And this job is important, not all tasks can be engineered thoroughly. But this job has its scaling limits.

A software engineer is not about coding per se, it's about designing software. It's all about designing the right code, not more code. Work smarter, not harder, for scale. You ask them to write FFT and they'll find a way to depend on it from a common implementation so they don't have to maintain an independent implementation. I've personally found LLMs very bad at this type of work, the same way you and others relying to you describe it. (Ok, maybe FFT is overly simple, I'm sure an LLM can import that for you. But you get the idea.) LLMs have statistical confidence, not intellectual confidence. But software engineering generally works with code too complex for pure statistical confidence.

No offense to the LLM fans here, but I strongly suspect most of them are closer to the programmer category of work. An important job, but one more easily automated away by LLMs (or better software engineering long-term). And we can see this by how a lot of programming has been outsourced for decades to cheap labor in third-world countries: it's a simpler type of job. That plus the people biased because their jobs and egos depend on LLMs succeeding.

> The consensus on HN seems to be that you can be pretty productive with LLMs as coding assistants

If you define "productive" as writing a simple CRUD web application that your 13-year-old cousin could write between two gaming sessions, then you'll consider LLMs as sacred monsters.

Snake oil vendors always had great appeal over people who didn't know better.

That’s very literally “productive “.

I have a similar experience, but I still use LLMs, just a bit differently. I pretty much never ask it to generate complex code. I also rarely ask for definitions or facts, cause of the tendency to generate garbage answers. What I use it for is usually tedious stuff that is easy to do, but would take me more time to type, rather than ask the LLM.

For example:

* I need a simple bash script for file manipulation or some simple tasks like setting up a project (example: download a secret from AWS SSM, check if an executable exist, if it doesn't write instructions on how to install it on most popular systems etc)

* I need a simple HTTP API, nothing fancy, maybe some simple database usage, maybe running some commands, simple error handling

* I need a YAML file for Kubernetes. I say what I need and usually, it gets most of it right

* I want an Ansible task for some simple thing. Ansible is quite verbose, so it's often saving me time

* I have a Kubernetes YAML file, but I want to manage it in terraform - I'll then ask to convert YAML to a terraform entry (and in general converting between formats is nice, cause even if you have only a piece of what you want to convert, LLMs will most of the time get it right)

* Surprisingly, it often gets openssl and ffmpeg commands right - something I always have to google anyway, especially openssl certificates generation or manipulation

* Giving it a function I wrote and writing test samples after providing a list of what it should test (and asking if it can come up with more, but sadly it rarely does generate anything useful on top of what I suggest)

> I gave GPT two small twists so it's not a simple case of copy/paste

Why? I see it like querying a database of human knowledge. I wouldn't expect a SQL database to infer information it's never seen before, why would I expect an LLM to do so?

I use it where I know a solution exists but I'm stumped on the syntax or how to implement it in an unfamiliar environment, or I want to know what could have caused a bug based on others' experience etc.

I stop it from coding, just ask it to brainstorm with me on a design or plan. After a few iterations it knows what I want and then I ask it for specific outputs. Or I code those myself, or some combination.

I’ve found that it’s sometimes amazing and sometimes wastes a lot of my time. A few times it’s really come up with a good insight I hadn’t considered because the conversation has woken up some non-obvious combination. I use ChatGPT, Claude, Perplexity and one or two IDE tools.

what did it do when you told it all of those things? was it able to fix the problems when you pointed them out? did you give it one prompt and expect perfect code out on the first try? is that how you code? all your code complies and runs flawlessly first try? I'm jealous. it usually takes me a bunch of passes before I get things right.

here's a chat for a uc and LCD chip that I picked at random (and got the name wrong for) (and didn't want raspberry pi code for so it stopped it short on that response)

https://chatgpt.com/share/2004ac32-b08b-43d7-b762-91543d656a...

Do you find that you accept GitHub copilits line completions like 90% of the time?

It's the same for me. It has to work 100%. I have 15+ years of coding experience, so I can code arguably fast. If I ask the LLM to code something, and it only partially works, then debugging and fixing the generated not-so-optimal code takes me more time than simply writing it from scratch.

There are also some more gotchas, like the generated code using a slightly different package versions than the installed ones.

If your productivity is near-null, LLMs can make you 100x more productive.

Right now AI is like a chissel. It's a very useful tool, but not useful for everything. Banging your head against the wall of capabilities will give you an intuition when you will pull this tool. Just like you learned how to use a search engine effectively over the last 20 years.

When you are familiar with LLMs, then a question from someone who doesn't use AI is very obvious. It's the same feeling you have when you roll your eyes and say "you could have googled that in 10 seconds".

It's either explaining code where you don't even know the lingo for or what the question could be. Or touching code with a framework you never used. Or tedious tasks like convert parts of text into code or json. Or sometimes your mind is stuck or drifts off. Ask AI for an idea to get the ball rolling again.

Yes, discovering what works and what doesn't is tedious and slower then "just doing it yourself". Like switching IDEs. But if you found a handful of usecases that solve your problems, it is very refreshing.

The most valuable thing AI has done for me with coding is commenting my code and creating documentation from it. It saves so much time doing a task that hardly any of us actually want to do - and it does it well.

This is just Dunning-Kruger effect.

I think tools like Copilot are best for people learning a popular programming language from scratch. You get better answers and easier problems to solve. The more experience you have the more you can outpace it in terms of correctness and appropriateness.

Not a fan of Copilot - having a chat bot separate from your coding environment feels much cleaner.

I also find AI coding assistants useful, particularly when working on a framework heavy codebase (Django, Spring etc.). While I don’t use it to generate massive blocks of code from descriptions just the mostly accurate tab completion of boilerplate code probably saves me 10-20% of my coding time. It’s not a groundbreaking advancement but certainly a decent optimisation.

How much do you end up paying for Claude + Cursor? I’m assuming cursor uses Claude api

100% this - Cursor's new inline multi-file editing with claude is staggeringly good. And it's _so_ new still.

It’s just a souped up version of Aider with an easier interface, isn’t it?

I have the same experience; I hope some open source/weights model gets to the same standard as I find a company like Anthropic having too much power to just shut out parts of the world (it did in the beginning when not available in the EU so I had to use proxies etc).

It is of vital importance (imho) to get open models at the same level before another jump comes (if it comes of course, maybe another winter, but at least we'll have something I use every day/all day; so not all hype I think).

BigCos won't let their employees use these tools to write code because you can't let code leak off prem. To keep from falling behind I use it on my weekend projects. I sense no urgency to remedy this inside, but I'm not sure how these tools would help with millions of lines of legacy code. They are awesome for creating new things or editing small things where you can build up context easily.

> BigCos won't let their employees use these tools to write code because you can't let code leak off prem.

Ridiculous blanket statement. A bunch of places use external LLMs.

I'm a dummy with very rudimentary script kiddie tier skills, and Claude has helped me do a bunch of personal projects that would normally take months of research and stackoverflow begging in a few days. For me the hype is real, but maybe not trillions $$$ worth of value real.

How did it change your workflow? I'm also a developer with 10+ years of experience. Generating code is not what I miss in my daily life. Understanding what you're doing and creating a mental model of it is the difficult part. Typing in the code is easy. I'm not sure how would a coding assistant help with that.

I feel like currently the group feeling the most impact from Claude is frontend devs as tools like v0.dev, Claude etc. can almost create frontend from API schema (I've tested only page per page) which is great. It's probably because there's examples on the internet and that's why it works well.

Here's the test for experienced engineers to see if the LLM hype is bullshit or not -- use cursor with Claude Sonnet. Build an app in a programming language you don't know at all. Don't stop prompting until you build it.

Here's a test for experienced engineers to see if the LLM hype is bs or not --

1. Use cursor with Claude Sonnet

2. Pick a programming language you don't know at all

3. Build an app in that language prompting only, don't stop prompting until you've run it successfully