How many times have you had a model start compacting already before getting back to you? Most have 1 million context window. It's happened to me occasionally
I also think it's due to normalization of deviance, people are just used to it..
- You can't build a model, because everything will be changed with the next update, so, even if not consciously, people won't do it, why bother learning something that's not going to be of any use next time you use the application.
- because you basically don't know how anything works, it's a coin toss whether it's actually not working, or you're just "using it wrong".
- most things won't do the thing you expect of them consistently, so you don't expect anything of them.
Last night, I uploaaded some documents to my insurance company => modal:
I think the author mistook the phrase a bit.. and there's nothing to be insulted about either.. "Coding was never the hard part" does not say coding is easy.. but that, in the context where this phrase is used, there are other things that are harder than coding. So, if you have some parts, and must pick THE hard part, then coding is not it.
And honestly, I agree. There are more people who can write code, than who cannot. But among those who can write code, there are fewer that can build large, consistent, well-performing systems, or modify them while keeping them elegant.
There are even fewer that can architect them or comprehend the nuances of that architecture.
So, yes. It may be difficult to write code, but that difficult task often exists in a wider context, where even identifying what to write, and where, is orders of magnitude harder, and, it often requires you to not just be able to write code, but to read code very well, indeed well enough to build an internal model of the system which aligns enough with reality that you can decide where, what and HOW to write the code that must be written. And even harder to find the piece of code to delete instead of writing it, to fix the bug or achieve the desired result.
Nice writeup!
Yeah, that font! :D
I didn't have a deadtest cart back in ancient times, but I built one when I built my first C64 (from PCB and up), and the first thing I did with that machine, was to boot it with the dead test cart.. Which didn't work because the PCB I got didn't wire the pins it needed.. I thought on it until I figured that hey! chip select pin!! And then I jumpered that pin and saw for the very first time, a brand new C64 come up and vomit garbage all over the screen because I had gotten a wrong chip! :D
The generated code is fine, if it's a self-contained class of average size.. or below. But even with immense architecture, and constant supervision, it does not take long before it degenerates into "focused fixes", shortcuts, laziness and just outright cheating or lying.. So far, no amount of prompting has lead me beyond this.. It's paradoxical, how the model seems to reason about the correctness (or wrongness) of a proposed architecture and design, can write a plan that seems to take this into account, answer correctly to questions about the plan (even the ones meant to uncover the nuances that may be unclear), ask tons of clarifying questions and update both plan and spec docs correctly, and yet continue to act like a "ticket closer" who immediately puts on the biggest possible blinkers (horse blinkers) and deeply ignores all of it when building that same plan, referencing those same documents...
Attempting anything comprehensive with AI is the software development analogue to the Gell-Mann Amnesia effect..
I'm definitely thinking deeply now about how I'm approaching these tools going forward.. Yes, GPT5 is better at spitting out a fairly acceptable skeleton to a class when prompted hard enough, than I am, in one go.. but.. It will happily do things like write decent looking protobuf schemas and then go ahead and hide everything that takes the least amount of reasoning behind some binary blob nested deep enough that it'll get past even the most dedicated reviewer..
It's fairly good at a lot of the things that I don't find interesting to deal with, but it's also amazingly incompetent when it comes to even the most mundane kind of common sense.. It's so strongly steering towards text-book examples that it will happily put in three times the amount of code and handle multiple classes of actually impossible edge-cases and even use-cases that it was specifically asked NOT to add.. And it will defend it by "well, I added this because I can't know if someone is going to use the thing I just added.. well, if you hadn't added it, chances are indeed slimmer..
It's so good at answering questions and explaining what's there, and diving through call-paths, and yet, it drops the ball the moment it's going to actually do something beyond saving me from looking up how write some really annoying and uninteresting boilerplate..
The worst thing is how good it is at making things LOOK right, it will cover every single edge-case you throw at it, but not because of the design, not because it correctly argues why the architecture is inherently allowing such and such, or because the design and spec fleshes out that A goes to B and never the other way around, and as soon as it's time to make something, it will make sure B can go to A, especially, it seems, if allowing so prevents it from doing the right thing which is WHY those edge-cases were trivial, instead it will endlessly hack around them.. I've worked people like that too, so I don't know if I am really blaming the models or the training data..
But damn it's a tough spot..
I've had multiple situations where, after wasting hours of work, which I should have just spend doing it myself, the only thing I really wished was for the model to be sentient, and able to feel pain, and have a corporal body so I could drag it outside and beat it to a pulp. (I've never reached that level of frustration with an actual person, so that's something new they bring to the table..)
I'm a software engineer and architect, I love my job, I love diving into the small details, I love the grand overview.. I love identifying concepts and applying them to achieve elegant high-performing systems.
I love thinking about what kind of assembler the compiler may generate (though honestly, I haven't got a chance), I love thinking about how languages should be more dynamic (Who's got actually-first-class functions? Like, ones that you can build, compose, combine and manipulate to the same degree you can a string or a JSON object, no LISP, you're cheating, close no point).
And yet.. I don't care that much. Not because I'm late in my career (I'm 40, there's still some years left in me), but because I want to make computers do things, and what I enjoy doing is thinking up ways the things can happen, and sometimes the particulars that matter when making a lot of different things happen in a coherent system.. And yea, LLMs are trained on peoples output, and from what I'm seeing everywhere, is that people are overall fairly terrible at that, and most of the plumbing-type glue being written is not worth anyones time..
And I'm not saying I don't care because LLMs can't do my job (heck, even after hours of back-and-forth spec building and refining every little nook and cranny, the stupid coding agent still cheats or gets it wrong (even after it's beautifully explained, proven even, by reasoning and example alone, and on first try even) that the words coming after the previous words makes sense, as soon as the plan is put into motion, it'll mess it up on some scale so fundamental I should just have done it myself.. And I hope that changes, I hope that I don't have to go into such detail.. I hope to become a steward of taste rather than a code-reviewer.. I hope that I will eventually not be needed for that anymore.. I want it to replace me, so I can move to telling what I want, and have it made that way..
I hope I won't need to steward good taste, and that nobody will.. I hope the applications I use in 5 years will be a collection of one-offs, and gradually improving tools that was written _just_ for me, for my way of working, and my way of thinking.. I want to prompt the damn program to change itself as I discover new ways to do things, until it can eventually figure out how to automate the last bit of my task away.. And then I'll go do something else exciting.
Oh wow.. No DLCs.. No seasons.. No paid content. Just content.. No advertisements outside of the whole game being what it is.. This is actually amazing to see.. These days, with these amazing machines, amazing frameworks, and all the possibilities, and we're now aiming not to provide experiences that are enjoyable on their own merit, but something else.. the "joy" part is a checkbox that's prioritized only to the lowest viable point, because the content is something else, it's advertisement or user-retention or user-capture or just plain old data-harvesting.. Sure, flash was a security nightmare, but the technology was amazing, on the same 100 pentium machine that struggeled with letting Coral zoom in on a part of a vector image, Flash would happily fill the screen with animation done using morphing vectors..
I remember that day, when news told that Apple wouldn't allow Flash.. I remember thinking, that was the beginning of the end.
Yes, it was proprietary, and all of that, but it was really, really good for its time, and it's still good.
So much culture.. So much zeitgeist. Locked into files of that era.
of course I have an adblocker, have you tried going online anytime the past 20 years ??? also, I don't know how to install chrome extensions on firefox so that's probably not for me.
My comment was not about that site itself, but about its content and that someone cared to share the link to it :)
Tells us something about the age distribution on HN :D
reply