Post

I think that may be the core division here: if you are an author or an artist the cons of generative AI very likely outweigh the pros for you If you are a software developer the opposite is true

5 reposts378 likes

I've been a professional SW dev for 24 years and no, not at all - it's just another tool in the box at best but it's not transformative and the findings about what long term usage does to brains makes me very wary of it

That’s harsh on software developers. Some of us can see beyond our keyboards, and externalities matter. Better division: if you’re a mindless consumer it’s all upside. If your role in society is to advance it with new knowledge, this world is a horror show.

The problem is that all of these products encourage you not to do that. There's simply no opportunity to get peer feedback on whether the code you've been given works or whether there might be better ways to do it. You're learning nothing in the process

Have you ever run against a study that changed a belief that you held deeply? Or do you dismiss anything you dislike without looking into it? Because while 18 minutes is enough time to read a paper, you seem to disagree with it while stating that if it were true that it wouldn't matter. Why is that?

the paper doesn't show that and the authors said so!

conputer dipshit@davidcrespo.bsky.social

they gave the caveats but here's my version: - experts on their own open source proj are not representative of most software dev - measuring time undervalues trading time for effort - tools are noticeably better than they were 6 months ago - it really does take months of use to get the hang of it

as a software developed I find that they improve some things (can be good for automated code review, though cant replace human review), are different but not better for others (writing code), can be much worse (enabling people to write too much code that others have to maintain and review)

This is only true if your acceptance criteria for art and writing are so superficial that it's impossible to have acceptance criteria for software that superficial. If you care about quality, all the same problems apply.

As a network engineer, I'm also against almost all forms of generative AI. I've yet to find a valid use case. Any maintenance plans/scripts some PM tries to generate "to save Chuckles some time" require more time/effort correcting than just writing it myself, ditto for meeting notes or RCA reports.

The thing that everyone near a data center can demonstrate on video from any faucet is the fake one, and you say I can read about it on a website that's famous for pushing neo-nazi advertisements? AI truly has done amazing things to the human mind.

🙋I've sucked at coding for over 20 years. So maybe that's why this year I've found some occasional uses for LLM generated code. I don't distribute my code to others (directly or in compiled form), so you don't have to worry about that.

Software has a fairly open culture of “stealing” from each other. Literature and Art have a culture of trying to hide the fact that we are all just repeating what we have seen before with slight modifications. Only when you become as successful as Picasso can they start to be honest about that

Software developers still get paid no matter how many copies they ship. Indeed, most developers work on software that will never "ship", their work is used only by their employer. It's a completely different economic relationship.

Lol collapsing all of human culture into “yeah it’s all derivative” what absolute bullshit from a singular minded STEM brain who never read a book that made them feel anything.

Ah, well that sounds pretty balanced. Tell me, are there any broader societal impacts of widespread AI that might help us settle this? Are there any potential risks? How about environmental concerns? Let’s sort this out together, shall we?

Zero results for search term self harm Zero results for suicide Zero results for porn ONE result climate change, and it just mindlessly parrots tech bro propaganda about per-search consumption instead of using data center projections by actual energy experts. Your collection speaks volumes

if you're a software developer your work is being stolen too. ffs the first thing that happened when they made github copilot available is it started spitting out things from private repos that it wasn't supposed to be training on.

Short term, perhaps. I think AI is an even bigger threat to developers' livelihood in the long run. Software is a means to an end, and I expect we'll all be redundant in due time. Art, by its nature, is a human expression. So there's got to be a cultural component as well, explaining the difference.

bro has never logged into GitHub to see 10 AI-generated PRs submitted by junior devs that didn't read anything before committing.

Also, there is a huge culture of sharing code already. Many passionate devs are FOSS advocates. While we haven't solved the "big corporations are making banks with the work we made for free", we are not shocked at the idea of giving it away or copying it.

Plus, while you may try to inject your soul into every part of a book, a program is likely made of a lot of boilerplate, and only a small part is really creative. For this, AI is ideal. You'll rarely see people laying bricks, saying it's a creative endeavor; most would gladly automate that.

hey simon just making sure you're aware armin is a racist who has expressed support for DHH bsky.app/profile/colo... i'm surprised to see a board member of the PSF in his replies like this given your recent widely-publicized rejection of NSF funding to avoid compromising your commitment to diversity

Punished Mace@colophon.bsky.social

Are you still massively pro-apartheid or have you moved more towards acceptance?

this is by no means the first such statement he has made which i'm sure you're aware of and i really think you should reconsider lending your public support regardless of how much you may agree with his statements. i deeply respect the PSF's decision and feel this undermines it.

I’m a photographer (&writer) uh it’s deeply mixed when you realize it’s an incremental change since like literally the 1970s The problem is mostly race to bottom + cost of viewing crap vs cost of making crap. But people like cheap crap!

Most of the generative AI discourse is two groups of people talking over each other, which is why there will never be reconciliation. Then you have some software developers saying “AI Agents can’t do anything” which is then extrapolated to represent all software engineers.

There is no benefit to AI period. FULL. STOP. The rest is just propaganda. LLMs dont lead to intelligence. This has been proven over and over. They lead to delirium. In humans and in LLMs.

Software developers work with a code assistant. Artists and other creative people are fighting a machine that figuratively digests their work and turns it into shitposts. LLMs are basically just digital colons, artistically speaking.

We asked a new-hire contractor to add a small feature. He checked it in, we later found a small bug. We asked him to fix the bug and also add in one other small thing. He did both in 1 commit (bad). The diff was 100%. We fired him. He was not hired to use AI to do his job.

Since it seems everyone on Bluesky is piling onto this post right now, I regret not having spent a few more moments considering my word choice for it Replace "the opposite is true" with "the opposite is likely true if you've figured out how to effectively use it"

You may now commence flaming me for suggesting that LLMs are useful for writing code or that they work better if you learn to use them - but take a look at the thread first since it's possible your points have been argued already

As the author of a few books (and a programmer), the trade-offs are in favour of AI. There's no money in 99% of book authorship. My writing is likely to have more reach and societal benefit as a result of being consumed, remixed and regurgitated by AI than being read my 10000 book-buyers.

I’ve tried lobbing some softball questions from my job (that don’t include any proprietary info) at LLMs and found them all to be hilariously confidently wrong. I have no doubt that LLMs are okay at rote work like mapping properties between objects, but that’s not revolutionary.

In Anathem, Neal Stephenson introduces Diax’s Rake (a parallel to Occam’s Razor) - it’s the yard implement your friends are to beat you with if they find you believing something only because you want it to be true. bsky.app/profile/josh...

Josh W. Comeau@joshwcomeau.com

Tools like Cursor or Claude Code make us *feel* more productive, but can we trust that perception? I added a new blurb to my 2025 post about AI, “The Post-Developer Era”, which digs into a recent study about developer productivity and AI:

I propose that if you are _actually_ creative, a "generative" AI will not help you generate anything good, because it will not be as creative as you, and will not have your experience in the craft. There are plenty of creative software developers.

I'm an author whose day job is in software and I'm thoroughly unimpressed with Gen AI. It hasn't come close to being the miraculous time-saver or labor-saver that it was hyped to be. I think the bubble is going to burst, and soon.

As someone in software the only benefit to LLMs I see is it will slowly destroy the pool of people with actual expertise in building and running software making me more valuable.

No, I'm not a software dev & see the benefits of being able to sift through large amounts of data in a fraction of the time to put it in a digestible format. You still have to gate keep AI to make sure it didn't go rogue, but it's helped me in my work for sure. Needs to stay out of the arts though.

I suppose the fundamental difference there is that you could tell it would work because you knew it hasn't been hallucinated out of whole cloth and other humans had replied beneath or upvoted to confirm it would work

Eh not so much, it was often people that had cludgy code and didn’t know if something bad was in it. It’s also where the AI probably pulls most its programming stuff from, so a lot of that bad stuff is in their also

Yeah, the difference is basically "do i want bad/kludgey code written by people and then discussed by other people to improve it?" vs "do it want bad/kludgey code poorly adapted by a bot with no context about any problems?"

If you are talking common patterns and algorithms then yes, I'd agree.. however, if you think thats anywhere close to the majority, I'd say whatever you are programming isn't very bespoke if thats your view of it.

We still are, not "were", copying and pasting code that works, that came w/ an explanation of why it works, and we did it with an understanding of why we wanted this code in our codebase. The teammate I had last year pasted 1000s of lines of chatGPT's output w/o reading or testing it. Not the same.

The devastating AWS outage was caused by an automated script, and only fixed by a real human overriding them. I intend to continue being one of those humans that can repair the damage done by robots...

It just feels too rushed. AI doesn't understand basic concepts like what day of the week it is (I'm not even kidding), but yet companies want to rely on it for multimillion dollar functions. There's a large disconnect there that doesn't seem to be looking to get solved anytime soon.

The big problem is AI (today) can't 'touch grass'. It can't discern what is real, and what is imaginary. It hallucinates. It can't reason from first principles, only from what it has been taught. Autocomplete on steroids.

Nope, it's negative there too if you spoke with actual developers and not solely people who cut corners with bullshit LLMs that are now fucking up core functions of the most widely-used operating system in the world.

The irony is it's attitudinal. AI isn't in a state to start having a really big impact on the writing or art markets. Still too slop But AI is a direct threat to software developers, not from replacing but by reducing demand through augmentation Still I'm like let's goooooooooo

Your critical thinking skills are atrophied from using generative AI to do your work for you. Your brain is a muscle and you need to keep using it. The petty comforts of your doomsday device are temporary, but the environmental, neurological and societal consequences will be forever.

I'm a software developer and I've yet to see AI coding that can handle something more complicated than what was already possible with a saved template or a stack overflow search. Similar to art, the people pushing it the most are people who don't code.

My take as a senior engineer with over a decade of experience... It won't make you a more knowledgeable software dev, but it is great for handling boilerplate code and other tasks that are low cognitive complexity/high effort. I'm not an AI/LLM "fan," but it does have specific use cases.

Yes, I have it turned on in VS Code and it's sometimes useful as a fancy autocomplete. I like how it tries to write code based on comments and comments based on code. But I would never put the LLM in charge of any project, and that's what people are trying to do with it.

as a programmer I do not accept tools that harm non-programmers just because they might be useful for me. That’s textbook sociopathic behavior, the type of thing that becomes normalized when programmers only talk to other programmers. It’s like saying “but asbestos is a good insulator”.

The anthropomorphized product that makes people emotionally addicted is despicable. The capabilities in the fields of NLP, OCR and code generation of the model architecture are disruptive and transform business processes. Seems people no longer are capable of having a nuanced opinion on any topic.

Most of the people who originally replied in this thread are career programmers for *decades* and not just riding the AI hypewave. Simon is on the board for the entire Python programming language and Armin created one of the most important Python libraries.

It's entirely possible, and even likely, for people with all of those qualifications to be complete morons outside their fields, and they are perfectly capable of being catastrophically wrong about certain things inside their field, or estimating second-order effects of their own work.

So are you saying that they are innovators turned misguided evangelists, or are you saying they are lazy grifters paid to publicly back a thievery machine?

The things that needed replacing were themselves embedded in strings - I needed to turn every [ into a " and every ] into a " but only only if they were inside a single or double or triple quoted string that represented a SQL create table statement That's not an easy regex!

AI provably makes people dumber. I may get cancer one day. I do NOT want my oncologist’s brain to be rotted with fucking ChatGPT. I may get dementia one day. Either give me the respect of a human caretaker or execute me. No fucking robot babysitter. Also, fuck AI art. It sucks

Simon's talking about the use of AI to write code If you think AI training data is stolen labor -- which I consider a strong argument! -- it's stolen, in this case, from software developers, who posted their code on sites like StackOverflow not expecting it to be sold to OpenAI & company

It does, yes And freedom zero, "you may use the software for any purpose," has prevented people from restricting use for things like military applications since the start of open source, yes We are now finding out why, beyond being a bad idea ethically, freedom zero is also bad for us programmers.

A properly Luddite response to code generating AI would involve poisoning the datasets... but also selling vetted datasets in ways that compensate the programmers fairly If I knew of anyone offering vetted AI, then I might be more open to learning to use it

AI does a terrible job with code. Yes, really. It's almost always inexperienced programmers touting it, and for small things. It hallucinates and makes errors at the same rate there too. This is just the latest boss's attempt to abuse that kind labor.

This is pure cope. But I guess my over a decade experience building actual novel things and not fucking with react in a tutorial makes me inexperienced, right? As opposed to someone who just doesn’t suck shit at using tools

It's worth remembering that programing is writing in languages, and the earliest post-punch card work was done by humanities majors. That same push to devalue labor deprioritized security and robustness which lead to the data theft issues and bloat we see now. The first enshittification.

There are may arguments against generative AI that I find highly credible - the environmental impact, the fact that the training data is unlicensed, many negative impacts of how it is used, AI psychosis and other harms to users I don't think the idea that it's terrible at code holds up any more

Uhhh no - all of the people who are going through school now and relying on ai will lack the critical thinking skills required to build good products. Anecdotally I’m already seeing the over reliance on ai with some devs i work with and how dumb it has made them.

The only time it could be remotely helpful is if you already know what you're doing and you treat it like a single page stackoverflow search but you still have to check it and I'm totally fine with just stackoverflow search (pre-all this mess because it's fucking everywhere now)

While it's slightly slower up front because you have to debug and correct the LLM output to get it functional, it has the upside of preventing gaining in depth understanding over time thus making each subsequent use less and less efficient.