If Codeberg is trying to "compete" against GitHub and GitLab, why does it refuse to take a look at AI assistants? Apart from infringing on authors' rights and questionable output quality, we think that the current hype wave led by major companies will leave a climate disaster in its wake: https://disconnect.blog/generative-ai-is-a-climate-disaster/

Other _sustainable_ (and cheaper!) ways for increasing efficiency in software development exist: In-project communication, powerful automation pipelines and reducing boilerplate.

Generative AI is a climate disaster

Tech companies are abandoning emissions pledges to chase AI market share

Disconnect

@Codeberg I fully agree and add that »AI« is intended, created and developed (further) to be a surveillance, manipulation, control, and oppression network in every net IT application far beyond the obvious quality and job killer features. Rainer Mühlhoff made clear in a C3 event not long ago that »AI« is a vast global data collection project failing the check against (almost) every social criterion in the United Nations sustainability charta. He has analysed our (the majority's) everyday’s IT routines, describes the mobile IT world and, more and more, PC net applications as feedback loops to collect training and socioeconomic data for GAFAM AI development and Big Data stores.

KI – Macht – Ungleichheit. https://media.ccc.de/v/ce4743cc-50ad-4597-bcc8-58e1a7e53c20

KI – Macht – Ungleichheit.

KI beruht auf der weltweiten Ausbeutung nicht nur natürlicher, sondern auch sozialer Ressourcen. Um KI nachhaltig zu gestalten, müssen wi...

media.ccc.de

@mupan @Codeberg well said.
>Rainer Mühlhoff made clear in a C3 event not long ago that »AI« is a vast global data collection project
im interested to see this.... is it on fedi or peertube?

>feedback loops to collect training and socioeconomic data for GAFAM AI development and Big Data stores
have you heard the the acronym CAGEMAFIA(includes cloudflare, e.muskrat, intel, akamai)?

@frogzone @Codeberg I didn’t know the acronym CAGEMAFIA before. Thank you.

It’s almost in Fediverse, it’s on @mediacccde . Easiest browsed and searched with @newpipe. Since many asked (I’m happy!), I added the link in the original toot.

Thank you for liking my phrasing.

@mupan @Codeberg thanks i'll look it up!

> I added the link in the original toot
interesting.... is that the toot starting with "I fully agree and add that »AI« is intended...."? the link isnt not showing up in ur toot still, maybe the edit isnt propagating but need to confirm that we're talking bout the same toot. :)

>Thank you
its my pleasure to help ppl fight the tecnofash :)

@frogzone @Codeberg

Yes, https://digitalcourage.social/@mupan/112893587229279358

mupan - @Codeberg I fully agree and add that »AI« is intended, created and developed (further) to be a surveillance, manipulation, control, and oppression network in every net IT application far beyond the obvious quality and job killer features. Rainer Mühlhoff made clear in a C3 event not long ago that »AI« is a vast global data collection project failing the check against (almost) every social criterion in the United Nations sustainability charta. He has analysed our (the majority's) everyday’s IT routines, describes the mobile IT world and, more and more, PC net applications as feedback loops to collect training and socioeconomic data for GAFAM AI development and Big Data stores.

KI – Macht – Ungleichheit. https://media.ccc.de/v/ce4743cc-50ad-4597-bcc8-58e1a7e53c20

mupan 📚 (@[email protected])

@[email protected] I fully agree and add that »AI« is intended, created and developed (further) to be a surveillance, manipulation, control, and oppression network in every net IT application far beyond the obvious quality and job killer features. Rainer Mühlhoff made clear in a C3 event not long ago that »AI« is a vast global data collection project failing the check against (almost) every social criterion in the United Nations sustainability charta. He has analysed our (the majority's) everyday’s IT routines, describes the mobile IT world and, more and more, PC net applications as feedback loops to collect training and socioeconomic data for GAFAM AI development and Big Data stores. KI – Macht – Ungleichheit. https://media.ccc.de/v/ce4743cc-50ad-4597-bcc8-58e1a7e53c20

digitalcourage.social

@Codeberg

Thank you Codeberg, glad to hear you aren't falling for the hype! 👏

@Codeberg One argument I have against AI in coding is that it reduces the knowledge you obtain from programming it yourself.

If you have an issue to solve, you may first try a few ways and if that doesn't work, you research the issue to see if someone else has had the same issue and found a solution (i.e. by looking at #StackOverflow) and if you're lucky, you implement the solution you found, else you communicate with others who have experience in it to find a solution.

Meanwhile with AI, you tell it the issue and it spits out a copyable code, often times with a bare-bones explanation that lacks info and context...

It may help beginners getting into programming, but it reduces the overall learning experience you may have from usual trial and error, and researching a solution.

@andre_601

> "Meanwhile with AI, you tell it the issue and it spits out a copyable code, often times with a bare-bones explanation that lacks info and context..."

You get something that might look like code, and with some luck does what it is supposed to do.
But without understanding the code you cannot trust it.
Does it actually do what it should do?
Does it even more? Maybe something unwanted/stupid/dangerous/...?
Is the code free to use in my OSS project? ...in production?

@Codeberg

@andre_601 @Codeberg A lot of the time too, I learn more from my first exploratory failings than from the actually correct one :)
@andre_601 @Codeberg ...and here I am typing out code snippets by hand instead of copy-pasting them 😅

@Codeberg
> Other _sustainable_ (and cheaper!) ways for increasing efficiency in software development exist

This is the big thing for me even if I ignore everything else that's an issue. I've tried to use AI code assistants multiple times and I can't shake the feeling that getting good with them would be a lot of work for not nearly as much payoff, when I know there are infrastructure things I could work on that would make me WAY more pawductive and that wouldn't have those downsides >w<

@Codeberg Even with a local assistant, there are better things I can do with that processing power.

And I can't shake the feeling even in the instances I've tried one and found it helpful, that the stuff it's helping with is kind of bullshit. If I'm going to throw so much compute at something, at least it could do something more significant than speed up boilerplate.

@Codeberg One illustrative moment for me was OpenAI talking about setting up dedicated VMs for their models to run code in to give you the result.

And my immediate thought was, wait a sec, *I* don't have dedicated VMs to run code in. Why not? Wouldn't that be more useful for me? Why does the LLM have better resources than me?

@foxyoreos @Codeberg i tried copilot during the free technical preview, and some time more recently tried a copilot-like that was self hosted and ran on my own GPU (copilot would have been a lot more effective but i wasn't gonna give them money just to try it)

and quite honestly, i just found it more distracting than useful, it just made me less effective

when i had it i'd often end up waiting for it, when more often than not, it honestly would have been faster to just write it myself, instead of waiting for it to generate, and then reading it to make sure its actually what i want, and potentially tabbing through alternatives or trying to prompt it if its not what i want...

a lot of the time i already knew what i'm trying to do and it was just a distraction, sometimes i'm trying to do something new (eg the reason i'm writing it myself in the first place is that i'm doing something existing solutions don't) it would fail to grasp that be actively unhelpful (by clinging to assumptions from the more common method that i was trying to deviate from), and in the times i actually don't know what i'm doing (the situation "help" in theory could be most applicable) my problem is usually more big picture than what a code autocomplete AI is going to be any help with (and even if it could, usually my struggle is more with understanding the problem and wrapping my brain around it than actually implementing it, even if i wanted to make the magic box barf out an implementation i'd still need to understand the situation enough to explain it to the magic box and if i understand enough for that i understand enough to start doing it myself)

despite being someone with quite bad motivation who you'd think could benefit from the "help" i found it just wasn't helpful, i think i'd benefit more from improved code search/navigation and documentation

sure maybe its just that i'm not using the AI "correctly" but for the effort of learning to use AI more effectively i could also just already be on the way of just doing it myself, and honestly if i'm programming, i'd rather be programming not trying to understand and review code generated by a magic box (has a much lower energy consumption to be doing it without the magic box too)

@delta @Codeberg

> i think i'd benefit more from improved code search/navigation and documentation

This, I have this thought almost every time I try to use one. Maybe I could get better, but..

> sure maybe its just that i'm not using the AI "correctly" but for the effort of learning to use AI more effectively...

Yep, that's the catch. I'm supposed to invest a lot of time into learning something that's constantly changing. There's other stuff to learn.

@delta @Codeberg I feel weird saying that because I know it's a trope to say "the thing I don't like also isn't as good as everypony says", but literally I don't think I've had a coding session where an LLM was more than neutral. Could just be me, I'm not saying nopony at all could find them helpful, but like..

There are a lot of things I could do to improve my code, I don't see why I should give LLMs a special investment.

@delta @Codeberg Even ignoring all the other stuff, idk

It just feels especially weird to look at the energy cost and be left thinking, wait a sec, we're going to destroy the planet for *this*?

@delta @Codeberg

> despite being someone with quite bad motivation who you'd think could benefit from the "help" i found it just wasn't helpful

Also lol I get what you mean and that's a really good point. I could benefit a lot from somepony just getting me to write code, bouncing ideas off of them. Even if it wasn't good at writing code, at least it could do that!

And it's surprisingly bad at doing that in a lot of cases X3

@delta @Codeberg @foxyoreos like you said, the main part about coding is understanding the big picture and what to do next. I feel like rather than telling the AI code assistant to do something and coaxing it to get the right result, it'd be simpler if I removed the middleman (middlebot?) and wrote it myself ✒️

The most important productivity boost I gave myself is touch typing. It's still paying off ⌨️

@Codeberg Aside from these valid points, there's currently no obvious hindrance to using Forgejo as a source and/or a destination for generative AI. It doesn't need anything more to be baked in or tacked on for it to be useful for supporting AI or ML, as it is.

@Codeberg Yeah in my day-to-day I find I don't need help writing sloppy code, I need help writing clean code!

A *well-designed* static type system is much more helpful...

Though a dynamic one is also useful in the right context...

@Codeberg so AI IS an existential threat, just not in the way we expected. Cool 😭
@Codeberg Based on the AI-powered climate models, the best way to virtue signal your programming is to use a magnetic needle on a powered-off HDD, whilst subsisting on dried beans.
@Codeberg I hadn't even heard of y'all until now, clearly need to check you out!
@Codeberg
Seriously, thank you for that! 👍
@Codeberg it's good to keep things simple 👍
@Codeberg I'm glad Codeberg and Forgejo are not jumping on the AI bandwagon that most companies do because of FOMO.
@Codeberg Really glad the #enshitiffication helped bringing a light to those really "sensible" projects. No bullshit, just a great products with awesome people backing it up.  Thank you for your hard work.

@Codeberg

My top pinned post is the Cliff Note on this.

@Codeberg I'm really glad you've taken this position.
@Codeberg oh nice I’ve been wanting to find an alternative to GitHub; sad it was once amazing now even their enterprise support is just ai assistants. Gonna have a go!
@Codeberg and this is why you should use Goatbeg. With Goatberg AI you can code up to 55% faster while increasing your carbon footprint by the same amount! Now including free license violations.
@Codeberg Already moved my personal open source projects from GitHub to Codeberg, but thanks for demonstrating that I made the right decision!

@Codeberg

I would love to disable or remove any and all AI tools that seem to be built into more and more apps every day.

@Codeberg Amen. Reduce boilerplate, don't automate its generation in code that then needs to be maintained

@Codeberg

That's kind of like asking why, if GitHub/Lab punch you in the face repeatedly with rusty iron knuckles, why doesn't Codeberg choose to include this "feature", too.

@Codeberg TBH, not shoving "#AI" bs into your site is a really BIG bonus point!

I hate that shite...

@Codeberg For me Codeberg's competitive edge is that it doesn't have Artificial Insemination assistants.
@Codeberg The biggest "competitive advantage" is *not having* unwanted AI garbage all over the place, not mining your users' code to offer to others as plagiarism material.
@Codeberg More broadly, I see "not trying to be an IDE"/not trying to be "VS code web edition" as a huge competitive advantage. GitHub completely broke their UI for actual use (reading code, studying history, researching bugs) by making it into an awful klunky editor that's completely unusable on mobile.
@dalias @Codeberg the back button on PRs broke within a week of the microsoft buyout
@hipsterelectron @Codeberg The back button on GitHub is broken all over. Sometimes it does stuff like taking you back 2 pages but URL bar only back one, or vice versa.

@Codeberg Gosh, how *ever* can a code repo hosting company compete with another code repo hosting company without putting all their money and energy into a ludicrously expensive technology not directly related to hosting code repos, which doesn't even work, and which almost nobody wants?

It's like asking how the Yankees can compete with the Red Sox if they're not willing to eat all the footballs.

Good on you for making the sane call.

@Codeberg

Very weak argument imo.

How are the servers for codeberg powered?

@s_levi_s Codeberg is running a single server (~130W) on renewable energy. We know that we can reach more efficiency, mostly because we don't benefit from scaling effects yet.

The difference is more like this:
- we do not encourage users to use more resources, because it would cost us more. Most competitors do the opposite (e.g. recommending the use of heavy CI pipelines), simply because it makes more revenue
- we do not spend any computing power on tracking and unnecessary data collection

@s_levi_s
- we do not live in a careless "enterprice" environment: We reuse failing server SSDs for our private use, we know that others regularly replace just as a precaution
- we do not spend excess resources on high availability, due to a simple 80/20 rule: 80% gain with 20% effort. We know we could add more spare servers, more sophisticated backups, replace hardware more often to avoid potential failures ... but we don't. It's a waste of resources (incl. our donations).

~f

@Codeberg
So the main difference is your claim to operate more energy efficient and waste less resources?

You are still using resources... what is than the problem with an AI assistant created following your standards?

@s_levi_s Our original argument was that we should improve society somewhat. Yet we participate in it, however, that does not contradict what we say, even if we were not doing all of these things.

The problem with what you mentioned is that it is completely hypothetical (the article explains it), similarly to cryptocurrencies! And that even if it wasn't: Wasting hardware and AI scraping the entire Internet (not all of which is powered by renewable energy) 100 times is much less sustainable. ^n

@Codeberg

Well, than I could just argue that Codeberg is wasting hardware, and that the benefits for society are marginal compared to AI. Cost/reward. Codeberg is only a very small improvement to society.

That is also the difference to crypto imo... AI actually has use cases.

@s_levi_s @Codeberg I mean, I could also pull an argument and say that you can buy drugs with cryptocurrency and that this in itself is a use case that is much more useful than GitHub Copilot (not a serious argument), but I don't feel like playing the devil's advocate and arguing around in circles anymore...

This isn't going anywhere, so I might as well say that we can agree to disagree.

@n0toose Umm, who are you? Okay, we agree to disagree :D
@n0toose great job keeping your cool, I think I would have cracked and been a lot meaner 🙂
@s_levi_s Every little step counts, IMHO. I for one am very happy that I found @Codeberg a while ago and was able to move almost all of my Github presence over to a service that actually cares about the values I prefer (open source, transparency, non-profit, filled with smart and nice people).