Random Thread

[zompist]

I use a LLM for small tedious tasks: lately because I wanted to check a certificate/private key pair. Of course the idea isn't to solve the problem myself but rather to use the openssl command correctly. The LLM gave me the proper command to use; which I could have located in the doc myself or dug up in a security guy's blog somewhere.
I also used it to work out a hairy SQL request (again, nothing that wasn't in the doc already) Ironically the initial idea for SQL was to have something close to natural language!

Not to answer for Torco, but I expect his use case isn't that different. The actual math is probably already worked out, in a python library somewhere, you just save time on tedious Googling/Stack Overflow crawling.

I have no objection to your workflow here. I do have an architectural objection though: why are these libraries and tools so damn hard to use?

I know the answer, of course: they were designed by other coders, who understood their own tools and saw no need to provide extra tools on top to make it easier, or document them properly. Or they saw the need but no one gave them time to do it.

So, no one has time to write an API, and the result is that thousands of programs waste time solving the same problems over and over.

Though in my view the real job is in figuring out what exactly the client is trying to do, what he actually wants and what he really needs. All three questions have very distinct answers. Also the client has no idea of what he wants, is unaware of what he needs so teasing out the answer is close to psychoanalysis.

Absolutely true. And then there's the problem of the stuff they know, but don't know they know until they see your program doing it the wrong way.

[Ares Land]

I have no objection to your workflow here. I do have an architectural objection though: why are these libraries and tools so damn hard to use?

I know the answer, of course: they were designed by other coders, who understood their own tools and saw no need to provide extra tools on top to make it easier, or document them properly. Or they saw the need but no one gave them time to do it.

So, no one has time to write an API, and the result is that thousands of programs waste time solving the same problems over and over.

Just so. Also, ideally we'd have the time to peruse the docs carefully, and actually master the tools we use. In practice, there's often a manager somewhere throwing a hissy fit because he wants an answer yesterday.

[Raphael]

Wait, after years of arguing in circles about whether LLMs are going to destroy the world or not, there's now a serious, well-informed, and interesting discussion about what LLMs actually can or can not do in programming? Congratulations, everyone! You did it!

[Torco]

Some terrible non-AI news for a change: I spilled soup all over my chair and now it's ruined because the soup soaked into the upholstery. Honestly not sure what to do since I can't really afford a new chair, except maybe a cheap plastic one.

have you asked chatgpt lmao no but seriously i'd try spraying it with a hose, then using alcohol or lysoform or whatever, and finally letting it dry in the sun (it is summer where you're at, i hope). if you're looking to throw it away you might as well risk destroying it beforehand. and who knows, maybe the wood will survive it? [if it's wood in the first place, maybe its one of those gamer plastic-and-stainless contraptions, in which case rotting the textile is your only concern].

Some people actually like writing code, and conversely have little interest in bot-herding ─ and having an LLM write my code for me would take the fun out of it.

sure. and that's fine and good, but i thought we were talking about whether chatbots could actually produce usable code, not about fun. i'm saying they can, though mostly in relatively small snippets and with close human supervision: it still ends up taking a lot less worktime for a low-but-not-zero-competence-programmer to produce a tiny little applet, or automation, or useful scraper or whatever else, by using them.

I used to work for a company that did statistical programming, SPSS. And we had a team of statisticians who devised the formulas and checked the code to make sure it was correct. That's in addition to the designers and QA people, all separate from the developers. And believe me, there are a lot of gotchas in statistical programming. (And in what we consider simple arithmetic, as bradrn's link shows.)

I won't assume you didn't do testing... but what did it involve? Testing against a known piece of statistical programming? Running some calculations yourself? Eyeballing the code?

did you really work on spss? that's cool, i've sat in front of that thing for probably months of my life. a lot of these little stats thingies are just GUI implementations of already existing libraries, so I only do fairly superficial testing against other libraries, or against spss itself, or against previously known results where i have 'em. the only one i can remember where i wrote the maths themselves was one that did a particularly idiosyncratic thing for which there was no library. that one i tested a bit more thoroughly against the excel sheet they used to use it for, checking what it does with commas, points, nulls, strings, and tables formatted in weird ways, though i've found a couple minor glitches afterwards mostly to do with a function taking so long to process a large table that another function apparently starts running before the first one is done (i don't know man, javascript is weird, ended up adding a timer)

soo what Ares said, yea

I have no objection to your workflow here. I do have an architectural objection though: why are these libraries and tools so damn hard to use?

I know the answer, of course: they were designed by other coders, who understood their own tools and saw no need to provide extra tools on top to make it easier, or document them properly. Or they saw the need but no one gave them time to do it.

true but i might add: they were made by coders *for* other coders: your library can be exceptionally easy to use, and be better for it, but if it's still hidden behind "you have to install python and pip install my library and then after dependencies you can upload your data -not as a pandas df, you undertstand, but as a numpy array- and then you get to run the function" it's just going to be out of reach for like 90% of everyone. shouldn't be, but hey.

[Travis B.]

I don't really trust LLM's even for small tasks because they still are liable to hallucinate on them in addition to hallucinating on big tasks.

Take for instance the example I ran into of some LLM-generated code to start and wait for another process using Boost, a set of broad convenience libraries for C++. The LLM hallucinated that you should put the name of the executable in the argument list, a novice error that I wouldn't have made had I written the code myself. I ended up wasting half a day trying to figure out why the code didn't work when it would have worked right the first time had I spent the time to write it.

[malloc]

Wait, after years of arguing in circles about whether LLMs are going to destroy the world or not, there's now a serious, well-informed, and interesting discussion about what LLMs actually can or can not do in programming? Congratulations, everyone! You did it!

The problem is that if they can successfully take over programming, they can take over most other professions. Programming has a reputation as one of the more intellectually demanding trades after all. Also it strikes me as deeply unfair that programmers are now getting paid a quarter million a year just to bark orders at chatbots while artists and writers are losing their entire fields.

[Raphael]

How about a revived AI thread where everyone except malloc would be allowed to post? Or, for that matter, a revived Venting Thread in that vein?

Wait, after years of arguing in circles about whether LLMs are going to destroy the world or not, there's now a serious, well-informed, and interesting discussion about what LLMs actually can or can not do in programming? Congratulations, everyone! You did it!

The problem is that if they can successfully take over programming, they can take over most other professions. Programming has a reputation as one of the more intellectually demanding trades after all.

Once again, you fall into the mistake of assuming one single one-dimensional "intelligence", as if things are either "more intellectually demanding" or "less intellectually demanding", as if there couldn't possibly be different tasks that are intellectually demanding in different ways. What's more physically demanding - construction work or doing a track and field discipline at Olympic levels?

Also it strikes me as deeply unfair that programmers are now getting paid a quarter million a year just to bark orders at chatbots while artists and writers are losing their entire fields.

People have already pointed out the various falsehoods in that assumption to you.

[malloc]

How about a revived AI thread where everyone except malloc would be allowed to post? Or, for that matter, a revived Venting Thread in that vein?

Perhaps not. I am already treated worse than everyone else I know in real life and I don't see the benefit of extending that here. Also someone needs to provide a critical voice on AI and the tech industry more generally. Letting only techies discuss AI is like letting only tobacco executives discuss the merits of smoking.

[Ares Land]

I don't really trust LLM's even for small tasks because they still are liable to hallucinate on them in addition to hallucinating on big tasks.

They really don't hallucinate as much as they used to, but yeah I don't trust them either, you have to check (which not enough people do, granted.)
I'd compare them to a faster Stack Overflow -- you have to check Stack Overflow responses to.

Also it strikes me as deeply unfair that programmers are now getting paid a quarter million a year.

When I stop laughing, I'll have to call my former employers and ask them for the millions they owe in back pay; maybe they'd like a laugh too.

You hear tall tales all the time about enormous paychecks in the field, and some of them are even true.
For the vast majority of IT professionals though the pay is good but not nearly that good. The huge paychecks are for top managers and executives (and even then they're not 100% guaranteed.)

I also don't think LLMs have taken over programming. Did you read my post earlier, or zompist's? It's really not about writing code.

[Raphael]

Also someone needs to provide a critical voice on AI and the tech industry more generally. Letting only techies discuss AI is like letting only tobacco executives discuss the merits of smoking.

Nope - the smartest and most interesting criticisms of many things are often insider or former insider criticisms.

[malloc]

Nope - the smartest and most interesting criticisms of many things are often insider or former insider criticisms.

Except the techies clearly have a vested interest in promoting AI just as tobacco executives have a vested interest in promoting smoking. Meanwhile people outside the tech industry are losing job opportunities, dealing with the environmental consequences of data centers, and so forth.

It's worth noting that tech people overwhelmingly stress the power and (oddly enough the danger) of AI while skeptics of AI are overwhelmingly from outside the field.

[Raphael]

Nope - the smartest and most interesting criticisms of many things are often insider or former insider criticisms.

Except the techies clearly have a vested interest in promoting AI

No. A handful of tech industry bosses and VCs do. But trying to talk sense into you is, of course, a futile exercise.

[Raphael]

It's worth noting that tech people overwhelmingly stress the power and (oddly enough the danger) of AI while skeptics of AI are overwhelmingly from outside the field.

You're from outside the field, and you constantly stress the power and the danger of "AI". Except for rotting bones, everyone from inside, or formerly inside, the field here on the ZBB is sceptical of "AI".

[Travis B.]

It's worth noting that tech people overwhelmingly stress the power and (oddly enough the danger) of AI while skeptics of AI are overwhelmingly from outside the field.

You're from outside the field, and you constantly stress the power and the danger of "AI". Except for rotting bones, everyone from inside, or formerly inside, the field here on the ZBB is sceptical of "AI".

Yeah, I'm personally definitely within the "tech industry" yet I am very skeptical of "AI" as a whole (even though I do think it has its places, such as cleaning up noisy image data).

[Ares Land]

@malloc : you're understandably angry at mm, let’s call it, the Silicon Valley. But this board isn’t the Silicon Valley.
There are folks working in IT, former coders, several researchers who like many scientists can code, but not a single techbro.

Nobody is terribly hyped by AI either. Rotting bones comes closest but he's still very skeptical.

This board isn’t very representative sample though. IME many (not all, far from it) people working in what they call tech these days are very enthusiastic. (I'm trying to have a nuanced point of view on the Internet here so please help me: I said many people in IT; not 'all' not even 'most'); that is true. So are many people outside the field too, I think. We all know someone who can’t go buy groceries without asking ChatGPT.

Creative types do tend to be a lot more skeptical. I'd point out evryone here, btw are creative and techies.

My opinion on AI is disappointingly lukewarm. It’s a vaguely interesting tool, also a very annoying fad. If I see another
Nanobanana cutesy drawing or read another 'it’s not... it is' piece of ChatGPT slop or, gods help us, another "thinly" veiled Claude Code infomercial I'm going to puke.

I dearly wish that ten years from now I'll be able to point and laugh at all that. But nobody will even remember. We'll all be busy with the asteroid miner drones or bioengineered metatentacles moral panic.

[Torco]

malloc, try and add some class analysis into your thinking. obviously the techbro startup ceos don't have the same interest than actual tech workers who actually do the work have.

[Travis B.]

malloc, try and add some class analysis into your thinking. obviously the techbro startup ceos don't have the same interest than actual tech workers who actually do the work have.

But malloc just sees the "tech industry" as a monolithic whole, not as something comprised of many different people with different views and different interests.

[alice]

If I was managing a team of software developers, I would be very angry at anyone who used "AI" to write their code, especially if they didn't thoroughly test it. In virtually all the projects I've worked on you'd have needed a lot of prior knowledge and context before you could contribute anything of value¹. And the one time I've endured any kind of talk about how "AI" could benefit me in my work, I nearly barfed when I heard the word "productionise".²

Oh, and malloc, $250000 is a bit more than half of what I've earned as a software engineer in my entire working life. Everybody I've met in the industry who earned that in a year was in some sort of management position.

¹ Beyond writing simple toy applications, software development is actually really hard work, to which LLMs are much less suited than mundane things like building buildings.
² Or it might have been "productionize", since the speaker was probably from the USA.

[Raphael]

software development is actually really hard work, to which LLMs are much less suited than mundane things like building buildings.

I don't think I'd want to be in, or close to, a building built by LLMs. And I don't mean because of political principles.

[Ares Land]

software development is actually really hard work, to which LLMs are much less suited than mundane things like building buildings.

I don't think I'd want to be in, or close to, a building built by LLMs. And I don't mean because of political principles.

A friend of mine is a constryction engineer. Maybe I should ask. Maybe I don’t want to know - he works on nuclear reactors.

(If he does, he probably triple checks though. LLM or not, I wouldn’t go near any kind of building if they built them like we build software.)