10/10. No notes.
For all these people insisting you will be left behind for not using the tool, even if it is not magic:
How do I learn to read code? Because I will never blindly trust AI output, but I also do not know how to read it for correctness. Just how to create a test. I am beginner enough that writing is a lot easier than reading, and reading would honestly take me awhile.
Everything in me is straining against using AI and not having the skills to actually check its output, while knowing it sometimes spews bullshit that looks correct, is an actual legitimate barrier to using it and not just my personal distaste. Meanwhile I at least understand what I wrote. If I am to ever change my mind and unhappily jump on the train, feeling very very dirty but also not wanting to be left behind in a paradigm shift then I still have to be able to error-check it.
Yes I get the point of the article, but also there are some inventions that really did change the way we did things and probably had some people hyping it up as a must-have as per always—we just also have tons of examples of must-haves that did not turn out to be that way. And while you are being bombarded by the hype it is hard to know if the invention will fade away, will have a place but you can also get away without using it, or if it’ll be a thing everyone uses and that you’ll seem crazy not to, like refrigerators and the internet. Hindsight is 20/20, but in the present we’re walking around in heavy fog, possibly with blindfolds on.
I figure I’ll just keep not using AI and if I do get left behind, then I’ll force myself to use it. Learning to read code is useful either way ;)
I’m not defending AI here, but “people have been wrong about other things in the past” is a completely worthless argument in any circumstance. See: Heuristics that Almost Always Work.
I can’t help but read this while replacing “rock” with “large language model”
Heuristics that almost always work. Hmm.
Is it worthless to say “(the current iteration of) AI won’t be a huge revolution”. For sure, it might be, the next decade will determine that.
Is it worhtless to say that many companies are throwing massive amounts of money at it, and taking huge risks on it, while it clearly won’t deliver for them? I would say no, that is useful.
And in the end, that’s what this complaint seems like for me. The issue isn’t “AI might be the next big thing”, but “We need to do everything with AI right now”, and then in a couple of years when they see how bad the results are, and how it negatively impacted them, noone will have seen it coming…
Interesting article, but you have to be aware of the flipside: “people said flight was impossible”, “people said the earth didn’t revolve around the sun”, “people said the internet was a fad, and now people think AI is a fad”.
It’s cherry-picking. They’re taking the relatively rare examples of transformative technology and projecting that level of impact and prestige onto their new favoured fad.
And here’s the thing, the “information superhighway” was a fad that also happened to be an important technology.
Also the rock argument vanishes the moment anyone arrives with actual reasoning that goes beyond the heuristic. So here’s some actual reasoning:
GenAI is interesting, but it has zero fidelity. Information without fidelity is just noise, so a system that can’t solve the fidelity problem can’t do information work. Information work requires fidelity.
And “fidelity” is just a fancy way of saying “truth”, or maybe “meaning”. Even as conscious beings we haven’t really cracked that issue, and I don’t think you can make a machine that understands meaning without creating AGI.
Saying we can solve the fidelity problem is like Jules Verne in 1867 saying we could get to the moon with a cannon because of “what progress artillery science has made during the last few years”. We’re just not there yet, and until we are, the cannon might have some uses, but it’s not space technology.
Interestingly, artillery science had its role in getting us to the moon, but that was because it gave us the rotating workpiece lathe for making smooth bore holes, which gave us efficient steam engines, which gave us the industrial revolution. Verne didn’t know it, but that critical development had already happened nearly a century prior.
Cannons weren’t really a factor in space beyond that.Edit: actually metallurgy and solid fuel propellants were crucial for space too, and cannons had a lot to do with that as well.
Saying we can solve the fidelity problem is like Jules Verne in 1867 saying we could get to the moon with a cannon because of “what progress artillery science has made during the last few years”.
Do rockets count as artillery science? The first rockets basically served the same purpose as artillery, and were operated by the same army groups. The innovation was to attach the propellant to the explosive charge and have it explode gradually rather than suddenly. Even the shape of a rocket is a refinement of the shape of an artillery shell.
Verne wasn’t able to imagine artillery without the cannon barrel, but I’d argue he was right. It was basically “artillery science” that got humankind to the moon. The first “rocket artillery” were the V1 and V2 bombs. You could probably argue that the V1 wasn’t really artillery, and that’s fair, but also it wasn’t what the moon missions were based on. The moon missions were a refinement of the V2, which was a warhead delivered by launching something on a ballistic path.
As for generative AI, it doesn’t have zero fidelity, it just has relatively low fidelity. What makes that worse is that it’s trained to sound extremely confident, so people trust it when they shouldn’t.
Personally, I think it will take a very long time, if ever, before we get to the stage where “vibe coding” actually works well. OTOH, a more reasonable goal is a GenAI tool that you basically treat as an intern. You don’t trust it, you expect it to do bone-headed things frequently, but sometimes it can do grunt work for you. As long as you carefully check over its work, it might save you some time/effort. But, I’m not sure if that can be done at a price that makes sense. So far the GenAI companies are setting fire to money in the hope that there will eventually be a workable business model.
He proposed a moon cannon. The moon cannon was wrong, as wrong as thinking an LLM can have any fidelity whatsoever. That’s all that’s needed for my analogy to make the point I want to make. Whether rockets count as artillery or not really doesn’t change that.
Cannons are not rockets. LLMs are not thinking machines.
Being occasionally right like a stopped clock is not what “fidelity” means in this context. Fidelity implies some level of adherence to a model of the world, but the LLM simply has no model, so it has zero fidelity.
I feel this also misses something rather big. I find there’s a huge negative value of people I have to help through doing a task - I can usually just get it done at least 2x if not 5x or more faster and move on with life. At least with a good intern I can hope they’ll learn and eventually actually be able to be assigned tasks and I can ignore those most of the time. Current AI can’t learn that way for various reasons, some I think technical, some business model driven, whatever. It’s like always having the first day on the job intern to “help”.
The other problem is - unless I have 0 data security rules, there’s just so much the AI cannot know. Like I thought today I’d have Claude 3.7 thinking write me a bash script. I wanted it to query a system group and make sure the members of that group are in the current users .k5login. (Now, part of this is me not knowing how to prompt, but it’s also stuff a decent intern ought to be able to figure out.) One, it’s done a lot of code to work out what the realm is - this is useful generically, but is just code that could contain bugs when we know the realm and there’s only one it’ll ever operate in.
I also had to re-prompt because I realized it misunderstood me the first time, whereas I think an intern would have access to the e-mail context so would have known what I meant.
Though I will say it’s better than most scripters in that it actually does a lot of “safety” stuff we would find tedious and usually have to have something go wrong to add in, so … swings and roundabouts? It did save me time, assuming we all think it’s method is good enough - but this is also such a simple task that I think in some ways it’s barely above filling out a lot of boilerplate. It’s exactly the sort of thing I would have expected to see on stack overflow back in the day.
EDIT: I actually had a task that felt 100% AI could have done… if there was any way for it to know lots and lots of context. I had to basically fill out a long docx file with often AI like text describing local IT security standards, processes, responsibilities and delegations. Probably over 60% I had to “just make up” cause I didn’t have the context - for higher ups to eventually massage into a final form. But I literally cannot even upload the confidential blank form, forget about have some magic way for AI to get a brain dump from me about the last 10ish years of spoken knowledge and restricted wiki pages. Anything it could have made up mostly would have “been done” by the time I made a functional prompt.
I don’t think we solve this till we can run frontier models locally at prices less than a human salary, with integrations into everything a human in that position could access.
I can see this partly being true in that it’ll be part of a dev’s toolkit. The devs at my previous job loved using it to do busy work coding.
Oh god the hate in this sub. It is definitely another tool for a dev to use. Like autocomplete or a lot of other stuff a good IDE does to help you. If you don’t want to use it, fine. Perhaps you’re such a pro that you don’t need anything but a text editor. If you’re not, and you’re ignoring it for whatever petty reasons, you’ll probably fall behind all the devs who learned how to use it to get more productive (or, in developer terms, lazier)
Agreed. Like it or not, old school auto complete was the same thing, just not as advanced. That being said, comment op probably didn’t click the link.
“busy work coding” is that what you do when you try to look like you’re working (like a real dev)?
Real world development isn’t creating exciting apps all the time, it’s writing the same exact boring convention based code sticking to an established pattern.
It can be really boring and unchallenging to create your millionth respiratory, or you can prompt your ide to create a new repo and with one sentence it will create stub out 10 minutes worth of tedious prep work. It makes programming fun again.
In one prompt, it can look at my finished code and stub out half decent documentation that otherwise wouldn’t have been completed at. It does hallucinate sometimes, or it completely misunderstands the code, so you have to correct a few sentences, but the brain drain of coming to with the sentence structure to write useful documentation is completely lifted, and the code is now well documented.
AI programming is more than just vibe coding, and it’s way more useful than everyone here insists it’s not.
Feels like you should have automated it if you do it over and over.
We’re using it for closing security flaws identified by another tool. It’s boring, unchallenging work that is nonetheless still important. It’s also repetitive and uncreative enough that I’m comfortable having a machine do it.
There’s still human review but when it’s stuff like “your error messages should escape variables” or “write a longer function name” having a tool that can do most of the grunt work is valuable.
I agree that it will continue to be a useful tool. I’ve gotten a similar productivity boost using AI auto-complete as I did from regular auto-complete. It’s also pretty good at identifiying potential uses with code, again, a similar productivity boost as a good linter. The chatbot does make a good sounding board, especially when you don’t remember the name of the concept you are trying to implement or need to pro-con two solutions and you can’t find articles about it.
But all these claims of 10x improvements in development speed are horse shit. Yeah, you might be able to shit out a 5-10,000 LOC tutorial app in an hour or two with prompt engineering, but try implementing a feature in a 100,000 LOC codebase and it promptly shits the bed: hallucinating internal frameworks, microservices, ignoring internal practices, writing straight up non-functional code, etc. I’d you spend enough time prompting it, you can eventually massage the solution you need out of it; problem is, it took longer to do that than writing the damn thing yourself.
What a glorious site. I wish every webpage looked something like this
I’m skeptical of author’s credibility and vision of the future, if he has not even reached blink tag technology in his progress.
<blink>How dare they!</blink>
The future of web development is Angelfire.
Once both major world militaries and hobbists are using it, it’s jover. You can’t close Pandora’s Box. Whatever you want to call the current versions of “AI”, it’s only going to get better. Short of major world catastrophes, I expect it to drive not only technological advances but also energy/efficiency advances as well. The big internet conglomerates are already integrating it into search, and I fully expect within the next 5 years to have search transformed into an assistant-like chatbot (or something thereof).
I think it’s shortsighted not to see the potential of accumulating society’s knowledge and being able to present that to people in an understandable way.
I don’t expect it to happen overnight. I’m not expecting iRobot or Android levels of consciousness any time soon, but the world is progressing toward the automation of many things - driven by Capital(ism) - which is powerful in itself.
energy/efficiency advances
the potential of accumulating society’s knowledge and being able to present that to people in an understandable way.
We call this Wikipedia, please consider donating to keep it running!
As an old fart you can’t imagine how often I heard or read that.
I’d love to read a list of those instances/claims/tech
I imagine one of them was low-code/no-code?
/edit: I see such a list is what the posted link is about.
I’m surprised there’s not low-code/no-code in that list.
You’re right. It belongs on the list.
I was told several times that my programming career was ending, when the first low-code/no-code platforms released.
At my work we explored a low-code platform. It was not low on code at all. Beyond the simplest demos you had to code everything in javascript, but in a convoluted, intransparend, undocumented environment with a horrendous editing UI. Of course their marketing was something different than that.
That was not the early days of low-code mind you. It was rather recently; maybe three or four years ago.
“We’re gonna make a fully functioning e-commerce website with only this WYSIWYG site builder. See? No need to hire any devs!”
Several months later…
“Well that was a complete waste of time.”
You should click the link.
Hehe. Damn, absolutely fell for it. Nice 😂
Yeah but it’s different this time!
I do wonder about inventions that actually changed the world or the way people do things, and if there is a noticeable pattern that distinguishes them from inventions that came and went and got lost to history, or that did get adopted but do not have mass adoption. Hindsight is 20/20, but we live in the present and have to make our guesses about what will succeed and what will fail, and it would be nice to have better guesses.
Quality work will always need human craftsmanship
I’d wager that most revolutionary technologies are either those that expand human knowledge and understanding, and (to a lesser extent) those that increase replicability (like assembly lines)
It’s tricky, because there’s no hard definition for what it means to “change the world”, either. To me, it brings to mind technologies like the Internet, the telephone, aviation, or the steam engine. In those cases, it seems like the common thread is to enable us to do something that simply wasn’t possible before, and is also reliably useful.
To me, AI fails on both those points. It doesn’t really enable us to do anything new. We already had chat bots, we already had Photoshop, we already had search algorithms and auto complete. It can do some of those things a lot more quickly than older technologies, but until they solve the hallucination problems it doesn’t seem reliable enough to be consistently useful.
These things make it come off more as a potential incremental improvement that is still too early in it’s infancy, than as something truly revolutionary.
Well it’ll change the world by consuming a shit ton of electricity and using even more precious water to fill the data centres. So changing the world is correct in that regard.
It needs to be more trustworthy. If I have to double check everything, I still have to figure out how to do whatever it’s doing, then figure out how it’s doing the thing, then verify if it did it right. By then, I could have just done it in step 1.5 probably.
I do wonder about inventions that actually changed the world or the way people do things, and if there is a noticeable pattern that distinguishes them from inventions that came and went and got lost to history,
Cool thought experiment.
Comparing the first iPhone with the release of BlockChain is a pretty solid way to consider the differences.
We all knew that modern phones were going to be huge. We didn’t need tech bros to tell us to trust them about it. The usefulness was obvious.
After I got my first iPhone, I learned a new thing I could do with it - by word-of-mouth - pretty much every week for the first year.
Even so, Google supposedly under-estimated the demand for the first Android phones by almost a factor of 10x.
BlockChain works fine, but it’s not changing my daily routine every week.
AI is somewhere in between. I do frequently learn something new and cool that AI can do for me, from a peer. It’s not as impactful as my first pocket computer phone, but it’s still useful.
Even with the iPhone release, I was told “learn iPhone programming or I won’t have a job.” I actually did not learn iPhone programming, and I do still have a job. But I did need to learn some things about making code run on phones.
deleted by creator
I still think PWAs are a good idea instead of needing to download an app on your phone for every website. Like, for example, PWAs can easilly replace most banking apps, which are already just PWAs with added tracking.
They’re great for users, which is why Google and Apple are letting them die from lack of development so apps can make them money.
What the fuck is Silverlight
Microsoft Flash. Netflix used it for a while. I don’t remember anything else using it.
A bunch of Disney movie sites did for a while, back in the day when every movie had it’s own website with trailers, promo, and a link to buy tickets and/or the DVD release.
Ahh good times
The League of Legends launcher used it at one point. Not sure if it still does.
I was going to say there’s no way they still are since Silverlight was discontinued by Microsoft in 2013, but it is Riot Games so ¯\(ツ)/¯
Be glad, you never had to interact with that ‘technology’. I once did at an internship in 2016.
glorified autocomplete
I use it to discuss the pros and cons of potential refactorings, then laugh as it botches the implementation.
Which is honestly its best use case. That and occasionally asking it to generate a one-liner for a library call I don’t feel like looking up. Any significant generation tends to go off the rails fast.
You sir haven’t railed an entire ui out of your vibes up asshole
I’ve been using it to write unit tests, I still need to edit them to mock out some things and change a bit of logic here and there, but it saves me probably 50-75% of the time it used to take, just from not having to hand-write all that code.
Getting it to format documentation for you seems to work a treat. Nothing too complex, just “move this bit here, split that into points”.
If you use it basically like you’d use an intern or junior dev, it could be useful.
You wouldn’t allow them to check anything in themselves. You wouldn’t trust anything they did without carefully reading it over. You’d have to expect that they’d occasionally completely misunderstand the request. You’d treat them as someone completely lacking in common sense.
If, with all those caveats, you can get this assistance for free or nearly free, it might be worth it. But, right now, all the AI companies are basically setting money on fire to try to drive demand. If people had to pay enough that the AI companies were able to break even, it might be so expensive it was no longer worth it.
I use it to find easy to miss errors.
It pains me so much when I see my colleagues pay OpenAI to do programming assignments… they see it is faster to ask gpt, than learn it properly. Sadly, I can say nothing to them, or I would risk worsening relations with them.
You should probably click the link
I’m glad they do. This is going to generate so much work opportunities to undo their messes.
Except that they are research students in PhD course, it would exacerbate code messiness in research paper codebases.
Or open source projects…
I will use AI to prompt AI to code for me, free money 🤑
I left 10 years ago, web development is shit.
(Allow me to preach for a bit, I have to listen to my boss gushing about AI every meeting)
Compare AI tools: now vs 3 years ago. All those 2022 “Prompt engineer” courses are totally useless in 2025.
Extrapolate into the future and realize, that you’re not losing anything valuable by not learning AI tools today. The whole point of them is they don’t require any proficiency. It “just works”.
Instead focus on what makes you a good developer: understanding how things work, which solution is good for what problem, centering your divs.
Key skill is to be able to communicate your problem and requirements which turns out to be really hard.
It’s also a damn useful skill whether you’re working with AI or humans. Probably worth investing some effort into that regardless of what the future holds.
Though it’s more work with current AI at least compared to another team member - the AI cannot have access to a lot of context due to data security rules.
