This is just my personal opinion, but I think when a lot of people type a prompt for an app they want made and see the finished UI of it, they might think to themselves that "programmers are useless". But what they don't see or understand is the architecture of the app. They don't know what runs what. From redundant code to logic flaws and serious race conditions, AI won't really care. As long as it "works" and it compiles, then you have an app.
I work in electronics hardware development, where everything has to work on the first release, and the equivalent to compiling takes days to weeks. The field thrives on thorough and accurate documentation. If I want to incorporate a single transistor into a design, I can download a multi-page document detailing its behavior, and chances are that throughout an entire design, I won't come across any errors in any documentation I read.
Because the documentation is so good, a hardware design of a given complexity takes less time than the initial release of a software design with similar complexity, and it only needs to be updated when the design needs to accomplish something new.
Hardware design, especially semiconductor layout, has benefited from neural networks, and pretty much every computer processor released for decades has been partially designed by a neural network, but LLMs aren't having as big of an impact.
Because documentation for behaviors of software libraries and APIs isn't particularly thorough or accurate, and easy update processes have allowed a lax attitude toward getting software working correctly on release, an LLM's simultaneous access to the original documentation, as well as countless conversations about the difference between expected behavior and observed behavior, allows it to quickly create software routines more accurately than a human.
Tying all those routines together is still usually done better by a human, so a purely LLM written program isn't great, but a human programming while using an LLM can be faster and more accurate than a human alone or multiple humans.
AI is better than 99% of humans at coding. On the other hand, 99% of all humans don't code, at least not as a genuine skill. So the AI coding doomers are deeply concerned that all those people who didn't want to code in the first place, will stop coding. It's philanthropic.
Can we stop conflating AI hype and boosterism (i.e. LLMs are great and can only get better) with AI doomerism (i.e. AI research is gonna end up killing us all)?
They are two very different camps that because of ignorance or malice (Altman and Amodei trying to hype up their product) are lately being seen as one and the same, annihilating the body of philosophical research into long-term prospects for this technology and how best to contain it.
Yes. You need to hold its hand for it to produce anything good, and at that point I'd rather write it myself. It's useful for reviewing snippets, but just letting it loose on your codebase results in shit like windows 11, or 99% of websites made in the last year.
This is just my personal opinion, but I think when a lot of people type a prompt for an app they want made and see the finished UI of it, they might think to themselves that "programmers are useless". But what they don't see or understand is the architecture of the app. They don't know what runs what. From redundant code to logic flaws and serious race conditions, AI won't really care. As long as it "works" and it compiles, then you have an app.
Passing back and forth between different models to review the code is good for removing any junk or loose ends.
I realise a competent developer is going to do a much better job, but it’s not like the old days when every kilobyte of poorly written code counted.
I work in electronics hardware development, where everything has to work on the first release, and the equivalent to compiling takes days to weeks. The field thrives on thorough and accurate documentation. If I want to incorporate a single transistor into a design, I can download a multi-page document detailing its behavior, and chances are that throughout an entire design, I won't come across any errors in any documentation I read.
Because the documentation is so good, a hardware design of a given complexity takes less time than the initial release of a software design with similar complexity, and it only needs to be updated when the design needs to accomplish something new.
Hardware design, especially semiconductor layout, has benefited from neural networks, and pretty much every computer processor released for decades has been partially designed by a neural network, but LLMs aren't having as big of an impact.
Because documentation for behaviors of software libraries and APIs isn't particularly thorough or accurate, and easy update processes have allowed a lax attitude toward getting software working correctly on release, an LLM's simultaneous access to the original documentation, as well as countless conversations about the difference between expected behavior and observed behavior, allows it to quickly create software routines more accurately than a human.
Tying all those routines together is still usually done better by a human, so a purely LLM written program isn't great, but a human programming while using an LLM can be faster and more accurate than a human alone or multiple humans.
AI is better than 99% of humans at coding. On the other hand, 99% of all humans don't code, at least not as a genuine skill. So the AI coding doomers are deeply concerned that all those people who didn't want to code in the first place, will stop coding. It's philanthropic.
It’s the speed of writing the code that is essentially removing the need for traditional software developers.
When there’s inevitably a bug, Claude and Codex act possessed until they find it and fix it; usually well within 15 minutes with the right prompt.
Can we stop conflating AI hype and boosterism (i.e. LLMs are great and can only get better) with AI doomerism (i.e. AI research is gonna end up killing us all)?
They are two very different camps that because of ignorance or malice (Altman and Amodei trying to hype up their product) are lately being seen as one and the same, annihilating the body of philosophical research into long-term prospects for this technology and how best to contain it.
> Is everyone online just that bad at programming?
No, but the good programmers are hugely outnumbered by people whose skills are poor enough to be exceeded by a chatbot.
> Every vibe-coded software I've used has been ass, but these people swear that it's the best thing ever.
It is the best thing ever produced by them.
hype cycle and dysfunctional media, there are more rational takes out their
I recently discovered Frag, more nuance, less hyperbolics, some perspectives I've not heard elsewhere yet
https://www.youtube.com/@DataScienceatHome
Have you used it seriously?
Yes. You need to hold its hand for it to produce anything good, and at that point I'd rather write it myself. It's useful for reviewing snippets, but just letting it loose on your codebase results in shit like windows 11, or 99% of websites made in the last year.