How can they tell if you submit a patch that you produced with the help of AI though? They mentioned they found a bug in an AI-assisted patch. But an actual skilled programmer can easily produce code with the help of AI that doesn't have any bugs in it.
"Easily"? No. Just check the article, it's plenty of examples of skilled programmers stopping to use A"I" because long term they discover more bugs than they would have written themselves. And you can even write the code yourself and have an LLM check it because it'll find tons of false positives.
LLM are fundamentally probabilistic text generators. We need and expect tools that have an error rate as low as the hardware error rate.
What business have projects mandating the tool I use for my contributions? I’m responsible of the quality of what I send and should be the sole arbiter of if I want to clean generated code or not. This person is basically calling all their potential contributors too stupid to reasonably use some class of tools. That’s downright insulting.
> it's their project and they're entitled to set whatever rules they want
This is not the point.
Projects are free to ban vim users if they want and I’m free to say it’s utterly baffling and should be none of their business.
Rules don’t magically become acceptable because the people who put them in place were allowed to do so.
It reminds me of the opposition to static analysis I used to see sometimes a decade ago because people would be too dumb to sort out the real cases from the false positive. I’m always impressed by how change averse the field is.
> if you read the article this is is their primary concern
That’s my whole point.
They are basically implying that I’m too stupid to use some tools to produce quality code and they being the clever ones they are should shield us poor fools from ourselves.
That’s the insulting part. At least the copyright infringement angle makes sense.
I'm astounded that anyone even considered permitting LLM generated code in a browser engine, yet alone a technical committee. Why would you risk using a system that's known to generate logic errors and bugs in a system where security and correctness are the single highest priority?
When _writing code_, you achieve a certain level of understanding because you fundamentally need to make all the decisions yourself (and some of them are informed by factual statements).
When _reading code_, a lot of those decision points are easy to miss or take for granted which means you don't even notice there are alternatives. Furthermore, you don't look up the factual statements, therefore you have a lower level of understanding. Also you have no opportunity to review if those statements are actually true so decisions made on false assumptions get into the codebase.
Finally, reviewing code (to the same level of depth) is significantly more mentally taxing.
To be honest the whole premise AI code needs to be banned sounds like a bit of a histrionic caricature to me, so I might not be in the right mindset to accept this either, but, this does feel a bit histrionic too. Like, maybe in a vacuum, in a codebase, language, and functionality I'm not familiar with, or if I'm too inexperienced to be diligent, or I don't bother with tests...maybe I'm just old, and those things seem like necessary preconditions even though I'd merrily ignore them 12 years ago, and working on my own now predisposes me to being happy with the work
Whenever i open HN, there's a pro-A"I" post which makes me almost question if intellectual labor will be useful in a few years.
Yet when i try to use these tools, they fall short. They can probably be useful for generating half broken prototypes quickly because speed is more important that quality. But for real code, especially long term, they just never seem to end up breaking even.
This post somewhat restored my confidence in the world's sanity. There are still people who care about quality and who are willing to call bullshit bullshit.
EDIT: The most important part is that he links to tons of examples from other people. There's clearly a sizeable group of people who are not being heard often enough.
> Whenever i open HN, there's a pro-A"I" post which makes me almost question if intellectual labor will be useful in a few years.
HN is infested with grifters, sorry, entrepreneurs think that A"I" is their ticket to quick and easy riches
part of the confidence trick is keeping up the charade that A"I" is actually intelligent and not just a sophisticated bullshit generator
and if what you saw when you honestly tried using it last week was bad, then that's no longer state of the art, the latest model will fix everything, see, this latest benchmark they-definitely-didnt-cheat-on says so
I don't even see why "hackers" and "founders" should be related in any way. The hacker attitude (or approach to life) is that of curiosity and the need to build and create. The founder attitude AFAICT is how to get competent builders together and make money (on the positive end of the spectrum) or how to extract value from people instead of creating it (on the negative end of the spectrum).
I googled it and got "Hackers/Founders is a community of tech founders." I think that is sufficiently telling of what it's really about. It's another scam to "get those socially inept nerds to work for us and make us money without even realizing they are getting less than they deserve".
For me the biggest indicator of how much bullshit A"I" is, is how the A"I" tools themselves are bad. Visual Studio still doesn't have an "accept one suggested word" shortcut.
If your LLM is so great, why don't you make it add that feature, Microsoft?
How can they tell if you submit a patch that you produced with the help of AI though? They mentioned they found a bug in an AI-assisted patch. But an actual skilled programmer can easily produce code with the help of AI that doesn't have any bugs in it.
"Can" add in the chance if nonzero? Sure.
"Easily"? No. Just check the article, it's plenty of examples of skilled programmers stopping to use A"I" because long term they discover more bugs than they would have written themselves. And you can even write the code yourself and have an LLM check it because it'll find tons of false positives.
LLM are fundamentally probabilistic text generators. We need and expect tools that have an error rate as low as the hardware error rate.
I’m baffled by this.
What business have projects mandating the tool I use for my contributions? I’m responsible of the quality of what I send and should be the sole arbiter of if I want to clean generated code or not. This person is basically calling all their potential contributors too stupid to reasonably use some class of tools. That’s downright insulting.
I have lost all the good will I had for Servo.
> What business have projects mandating the tool I use for my contributions?
it's their project and they're entitled to set whatever rules they want
> I’m responsible of the quality of what I send
if you read the article this is is their primary concern
> I have lost all the good will I had for Servo.
I bet they're devastated
> it's their project and they're entitled to set whatever rules they want
This is not the point.
Projects are free to ban vim users if they want and I’m free to say it’s utterly baffling and should be none of their business.
Rules don’t magically become acceptable because the people who put them in place were allowed to do so.
It reminds me of the opposition to static analysis I used to see sometimes a decade ago because people would be too dumb to sort out the real cases from the false positive. I’m always impressed by how change averse the field is.
> if you read the article this is is their primary concern
That’s my whole point.
They are basically implying that I’m too stupid to use some tools to produce quality code and they being the clever ones they are should shield us poor fools from ourselves.
That’s the insulting part. At least the copyright infringement angle makes sense.
Aren’t you playing into the very drama that you are criticizing?
Is it disappointing that drama is an important reason people contribute to Servo?
While I agree with your point about being responsible for the quality of the code you send, please remember that...
> This post is my personal opinion, not necessarily representative of Servo or my colleagues at Igalia.
...so I don't believe it's fair to say that you lost all the good will you had for Servo based on just this article.
I'm astounded that anyone even considered permitting LLM generated code in a browser engine, yet alone a technical committee. Why would you risk using a system that's known to generate logic errors and bugs in a system where security and correctness are the single highest priority?
Frankly, I just assumed AI generated code would be treated with the same suspicion as human code and reviewed.
Reviewing code is harder than writing it.
When _writing code_, you achieve a certain level of understanding because you fundamentally need to make all the decisions yourself (and some of them are informed by factual statements).
When _reading code_, a lot of those decision points are easy to miss or take for granted which means you don't even notice there are alternatives. Furthermore, you don't look up the factual statements, therefore you have a lower level of understanding. Also you have no opportunity to review if those statements are actually true so decisions made on false assumptions get into the codebase.
Finally, reviewing code (to the same level of depth) is significantly more mentally taxing.
To be honest the whole premise AI code needs to be banned sounds like a bit of a histrionic caricature to me, so I might not be in the right mindset to accept this either, but, this does feel a bit histrionic too. Like, maybe in a vacuum, in a codebase, language, and functionality I'm not familiar with, or if I'm too inexperienced to be diligent, or I don't bother with tests...maybe I'm just old, and those things seem like necessary preconditions even though I'd merrily ignore them 12 years ago, and working on my own now predisposes me to being happy with the work
Whenever i open HN, there's a pro-A"I" post which makes me almost question if intellectual labor will be useful in a few years.
Yet when i try to use these tools, they fall short. They can probably be useful for generating half broken prototypes quickly because speed is more important that quality. But for real code, especially long term, they just never seem to end up breaking even.
This post somewhat restored my confidence in the world's sanity. There are still people who care about quality and who are willing to call bullshit bullshit.
EDIT: The most important part is that he links to tons of examples from other people. There's clearly a sizeable group of people who are not being heard often enough.
FYI the post is from she/her.
Oops, the one time I don't check, I end up getting it wrong, sorry.
Thanks for the apology, sounds like it was unintentional.
> Whenever i open HN, there's a pro-A"I" post which makes me almost question if intellectual labor will be useful in a few years.
HN is infested with grifters, sorry, entrepreneurs think that A"I" is their ticket to quick and easy riches
part of the confidence trick is keeping up the charade that A"I" is actually intelligent and not just a sophisticated bullshit generator
and if what you saw when you honestly tried using it last week was bad, then that's no longer state of the art, the latest model will fix everything, see, this latest benchmark they-definitely-didnt-cheat-on says so
Hackers and Founders meetup near me has become mostly just founders and they are all high on the LLM fumes. It's very weird.
I don't even see why "hackers" and "founders" should be related in any way. The hacker attitude (or approach to life) is that of curiosity and the need to build and create. The founder attitude AFAICT is how to get competent builders together and make money (on the positive end of the spectrum) or how to extract value from people instead of creating it (on the negative end of the spectrum).
I googled it and got "Hackers/Founders is a community of tech founders." I think that is sufficiently telling of what it's really about. It's another scam to "get those socially inept nerds to work for us and make us money without even realizing they are getting less than they deserve".
It's just a meet up man.
You go there, meet some people, have some conversations. Sharing information is the quintessential non zero sum game humans play.
For me the biggest indicator of how much bullshit A"I" is, is how the A"I" tools themselves are bad. Visual Studio still doesn't have an "accept one suggested word" shortcut.
If your LLM is so great, why don't you make it add that feature, Microsoft?