in reply to @pndc

@pndc
I also assume the ban will happen. But I give that ban one year. Then it will be released again. Because the ban will not solve the quality problems.

The true problem is that tests have no real
Meaning anymore, if LLMs can modify them. Tests coild to be separate repositories, which cannot be touched vy everyone. Or we could invent tools which check the grade of code modifications?

in reply to @pndc

@pndc
A tool like a linter which detects LLM code smells could be introduced. Which checks if SOLID and DRY was violated badly. This can never be perfect. But why can’t we find a way to automatically scan code upfront with abstract constraints.

Even humans can write really ugly code. Gosh, how often have I seen ugly code of beginners.

There must be a scientific way of maintaining the quality. Instead of „politics“ and „feelings“.

in reply to Liam Proven

I am asking myself what happens to the past LLM contribs. And cintribs which indirectly contain LLM code.

I personally find this debate hysteric. Because people will silently use offline LLMs for this and that and lie about the usage.

From my point of view the new situation is asking on how to do quality management in future. It’s asking about us needing new tools for admins. Better classifiers. Code Quality Scores.

How do you want guarantee a ban? This is just theory.

in reply to JackPearse

@jackpearse trust.

it's the same in many situations, like when you ask whether food or drink is vegan.

of course some asshole can lie — »yes« — and use a cow's breast milk instead of oat in the coffee.

then, if you find out they're lying, you out them and eject them.

letting people know that a harmful behaviour is not welcome goes a long way.

Questa voce è stata modificata (1 giorno fa)
in reply to Liam Proven

One day we will wake up and NOTHING will work anymore.

NOTHING.

but all the poor countries - the 'shithole' countries? - yeah, they will just continue to bring and sell and buy food at the local market, and ride their bikes back to their casa, and cook on their woodstove, after getting water from the well.

but we will all die within a week.

Questa voce è stata modificata (1 giorno fa)
in reply to Liam Proven

What I hate about this movement is: no one explains how you detect it? What about false positives?

I once contributed to OSS. My pull request was rejected. Because it had the „feeling“ of LLM. That feeling was me summarizing all changes of mine and a file listing. I wanted to safe the maintainers time. And exactly this „service“ was treated as „feeling“ like LLM made.

This is ridiculous. You work for days. And then people fuck you because they have a „feeling“.

Questa voce è stata modificata (1 giorno fa)
in reply to Liam Proven

But THAT is the issue. „You find someone did nor follow the rule“

HOW?

I have seen excellent LLM code. And I have seen ugly humand made code. And the ither way around.

How can you speak out a ban without telling how to control it?

This is what I am talking about. Anyways. I think this „so called“ ban will nit last for long. It’s missing a lot of issues in reality.

Questa voce è stata modificata (15 ore fa)
in reply to JackPearse

@jackpearse You are focusing on an implementation detail and not the purpose of the rule.

If it's fake, a fake person or some fake code, it will reveal itself. You don't need to test for it.

"So, John Smith, how does this patch work?"

(10 pages of markdown formatted wibble follows which doesn't answer the question.)

Or there's a bot config file. Or a credit line. Or it's a 200 line diff that changes 1 operator.

Bosh. Bot slop. Goodbye, "Mr Smith."

It's like the joke: "how do you spot a vegan?"

Answer: you don't. There's no need. They'll tell you.

In this case they will angrily complain, giving themselves away. Then they will stomp off and prompt the bot to create an entire distro, which they will never maintain because they don't know how.

It doesn't matter. You don't need to be able to test for it. Either they bitch about it (result, ban) or they cheat (ban) or they tell you (ban) or the bot tells you (ban).

in reply to Liam Proven

But it is already like this.
For me this ban is hysterical.

The OSS Community is popular for being very gate keeping. Never explaing stuff when people ask. Behaving elitarian to people who are new to i.e. linux.

From my perspektive of more than 30 years of IT experience, AI is just an amplifier of processes which already started with „python“ and „full stack development“.

And the „ban“ tells a story. It amplifies „gate keeping“.

This ban will make OSS fall back.

in reply to Liam Proven

I already addressed the points. Which you call „anger“.

This „ban“ is just tagging. And it sees a Problem like LLM. My point is: Not LLM is the issue but the culture. AI amplified it. And this ban makes it worse.

If you need to trust your team. Then you need to know your team. If you know your team, bans are obsolete.

If you don’t know the team, you need tools to detect bad members.

But everything is said now. Open Source is split into two camps now. And that is sad.

in reply to Liam Proven

We habe the same stupid procedures in the demoscene. You now have to sign „I have not used AI“ before uploading the contribution to the compo. How stupid is that? They added a layer of bureaucracy. WHY?

Instead of simply uploading shit without singning stuff. Since 1995 I am building software. And I have seen a lot of shit since the 1990ies. But never ever have I seen that IT became like politics.

From my personal perspective, the problem isn't the code. It's this debate.

in reply to Liam Proven

The real problem is „bullshit“ code. And instead of banning a particular Algorithm, I would go and search for ways to maintain code quality. At work I have seen truly ugly code, made by people. This is also a problem. The more coders participate the more helper functions appear. Also people produce Slop.

This could be a once in a lifetime chance to really advance IT quality management by improving code quality tools.

Instead we are closing borders.

in reply to Liam Proven

I am proposing automatic detection instead of a written „ban“. Just adding „this is banned“ will change nothing.

This is like trying to change climate change by apllying the rule where the caps on plastic bottles can no longer be removed.

The real problem isn‘t LLM. Microsoft owns github and is the bigest stakeholder of AI corporations. AI is indirectly the biggest OSS sponsor. (Like PostreSQL). The largest FOSS sponsors are AI-Corporations. Copilot is default in github.

in reply to Liam Proven

@ferrix @joe RedHat (or IBM, however you wish to look at it) is so incredibly all-in on LLMs it's not even funny.
redhat.com/en/about/press-rele…
redhat.com/en/blog/acceleratin…
"Red Hat believes that AI offers tremendous opportunities for open source projects and contributors. "
in reply to Joe Groff

@joe @Cal @arthfach @ferrix

Life is *way* too short.

FreeBSD can already run Docker containers. It's run Linux binaries for years. It only needs a little more wiring up of the bits that are already there to make it a single-click operation to just run one of the containerised Linux app formats (AppImage, Snap, Flatpak or something).

in reply to Liam Proven

@Cal @arthfach @ferrix my experience aligns with Liam’s that convincing FreeBSD to do desktop stuff is a very manual process still. that said, it uses the exact same drm drivers for Intel and AMD GPUs as Linux (albeit a few versions back), and ports contains a launcher for steam, so it’s possible. Depends on how aggressively proton and friends use cutting edge Linux kernel features that FreeBSD’s compatibility layer doesn’t emulate
in reply to Liam Proven

Thorvald says LLM is allowed to be used for the kernel. Then LLM use is banned by Debian.

Is the Kernel then banned?

In biology and medicine they use AI to detect cancer cells. Is this also banned then? Or are these type of programs allowed then?

What is with Computer Scientists developing free LLM tools. Also banned?

This kind of approach isn‘t realistic.

The overall issue is a quality problem. And needs tools to maintain quality. A ban is simply symbolic politics.

in reply to Liam Proven

debian has a ban on contributions in general, I have a few cool little command line utilities and programs I wanted to contribute and found out it would take a 3 year certification process where I had to travel around the world exchanging usb keys to be allowed to and would only succeed if , after it all, one of them was willing to 'bless' me.

I honestly have no idea how ANY of the packages in debian got there, but I can TOTALLY understand when projects host there own packages.

in reply to Rue Mohr

@RueNahcMohr The trouble with LLM code in a project like debian is simple, deep, and obvious.

If someone is using an LLM instead of learning, providing official feedback to the LLM instead of growing, and providing the LLM's reply to the feedback instead of engaging...

You're not training the next generation of developers. You're training a model.

in reply to Longplay Games

@Longplay_Games @RueNahcMohr you are missing the fact, that when learning yourself, tradcoding, expanding understanding, you also contribute to llms, even more i would say: your output is a compost for them and it makes them stronger. no license or ban will prevent it. until it's acked, there can't be an effective solution. and solution will be much more radical than Stallman was in the 80ties, i predict.
in reply to Rue Mohr

@Rue Mohr @Liam Proven

in case somebody is reading this and wondering, the process to start contributing packaging work to Debian is more or less:

* at first you need to do some work, submit it to mentors (a platform and a mailing list) and get a Debian developer to vet it: this can be quite easy or quite hard, depending on how niche what you're working on is: finding a volunteer that feels confident working on a python module with common applications is much easier than finding one for something written in a niche language that is only relevant for a few people. There may be a bit of back and forth requiring changes to the package: Debian covers a number of non-obvious usecases, and providing packages for that requires more nuances than building a package that just works on one computer.

* continue working on it in time: maintaining a package is a long term commitment, not something that works well with hit and run contributions. While doing so you're likely to establish a working relationship with one or more DDs.

* this is where being able to travel (or living in Germany :D ) helps: then you need to connect your contributions to a GPG (not USB) key and long-term identity: the easiest way is to meet one or more Debian developers in person, chat a bit, and exchange cryptographic signatures on that key, but since 2020-ish (for obvious reasons) there is at least one alternate procedure where meeting is optional. Having the same name on the GPG key and a government issued ID makes the process easier, but is not mandatory either.

* once you have both an history of contributions and a signature you may be given permission to upload a small number of packages without supervision (Debian Maintainer)

* and after some time as a Debian Maintainer you may, after a bit of exam, become Debian Developer and get permission to upload anything, voting rights, and everything.

Honestly 3 years between the first contribution and becoming DDs sounds even a short time, unless somebody has a *lot* of time to contribute, but it's not like your contributions aren't getting into Debian during that time, they only need an extra step or a few because they have to be supervised.

(Also, if your contributions aren't packaging related the process is different, and depends on what they contributions are related to. And there is a way for people to get voting (but not uplading) rights without ever having to learn how to package, if they are contributing in different areas.)

reshared this

in reply to Rue Mohr

@RueNahcMohr @dos

You were told earlier today, in this thread.

social.gl-como.it/display/3e3c…

What more do you want?


@Rue Mohr @Liam Proven

in case somebody is reading this and wondering, the process to start contributing packaging work to Debian is more or less:

* at first you need to do some work, submit it to mentors (a platform and a mailing list) and get a Debian developer to vet it: this can be quite easy or quite hard, depending on how niche what you're working on is: finding a volunteer that feels confident working on a python module with common applications is much easier than finding one for something written in a niche language that is only relevant for a few people. There may be a bit of back and forth requiring changes to the package: Debian covers a number of non-obvious usecases, and providing packages for that requires more nuances than building a package that just works on one computer.

* continue working on it in time: maintaining a package is a long term commitment, not something that works well with hit and run contributions. While doing so you're likely to establish a working relationship with one or more DDs.

* this is where being able to travel (or living in Germany :D ) helps: then you need to connect your contributions to a GPG (not USB) key and long-term identity: the easiest way is to meet one or more Debian developers in person, chat a bit, and exchange cryptographic signatures on that key, but since 2020-ish (for obvious reasons) there is at least one alternate procedure where meeting is optional. Having the same name on the GPG key and a government issued ID makes the process easier, but is not mandatory either.

* once you have both an history of contributions and a signature you may be given permission to upload a small number of packages without supervision (Debian Maintainer)

* and after some time as a Debian Maintainer you may, after a bit of exam, become Debian Developer and get permission to upload anything, voting rights, and everything.

Honestly 3 years between the first contribution and becoming DDs sounds even a short time, unless somebody has a *lot* of time to contribute, but it's not like your contributions aren't getting into Debian during that time, they only need an extra step or a few because they have to be supervised.

(Also, if your contributions aren't packaging related the process is different, and depends on what they contributions are related to. And there is a way for people to get voting (but not uplading) rights without ever having to learn how to package, if they are contributing in different areas.)


Questo sito utilizza cookie per riconosce gli utenti loggati e quelli che tornano a visitare. Proseguendo la navigazione su questo sito, accetti l'utilizzo di questi cookie.