Lately GitHub has been, well, lately,

not even lately. Uh, for the last like 2

years GitHub has just been being owned

in the news. Obviously, lots of

downtime, poor user experience, action

runners just

just the worst. Just general disdain is

what it feels like for the common user

of the platform. And so, alternatives

have been gaining steam. One of them is

being Codeberg. This is a platform

dedicated to the free and open source

software kind of environment /

community. And recently, they've made a

lot of news. Now, you're probably

asking, "Well, how did a small code

hosting platform make the news in such a

big way?" Well, it's because its members

have decided LLM projects, vibe coded

projects are no longer allowed on

Codeberg. Now, of course, this led to a

bunch of tweets, people just saying all

sorts of stuff, people complaining

saying this is absolutely a wrong move.

Some people even saying, "HEY, I'M

STOPPING MY donations to you, Codeberg.

Okay, how dare you?" But, of course, as

always, the internet latches onto some

sort of image and then claims this is

the truth when the the actual truth is a

little bit more complicated. So, I did

the extremely difficult job of actually

reading the blog right here with my

eyeballs, from left to right, top to

bottom, and I understood their

arguments. And so, I'm going to do my

best job to kind of talk about this, and

I think that they did some things really

well, and then other parts I think is

absolutely ridiculous, okay? Let's talk

about why Codeberg is banning mostly

vibe coded projects. Yes, I did use the

word mostly intentionally. Also, just a

quick sidebar, nobody complained when

they also, during the exact same

meetings,

disallowed cryptocurrency projects. In

fact, this is the direct ticket right

after banning LLM mostly LLM projects.

Yes, right here. You can't have projects

that harm the reputation of Codeberg,

such as relating to cryptocurrencies.

Dude, crypto finance bros just are in

shambles right now. Their boutique

artisanal code hosting platform

alternative to GitHub not welcome. But

before we do that, a quick thank you

from the sponsor.

>> Bros down.

>> WHAT HAPPENED?

>> I DON'T KNOW.

>> Did you push?

>> On a Friday?

>> Never.

>> We're going to figure this out.

>> How are WE GOING TO FIGURE THIS OUT?

>> GOT TO SWITCH YOUR BROS DOWN.

>> DO YOU GUYS NEED the wheel?

>> New VM detected.

>> It's sad.

>> I'm going to spend a couple hours

refactoring.

>> Off to the plugins.

>> Don't guess where your issues are. You

can see exactly where they are happening

with Sentry. Get all the context you

need to debug any problem because code

breaks, so fix it faster with Sentry.

So, I think the best place to start is

the terms of use. The terms of use was

updated right here. This is in section

two, paragraph seven. It says the

following, you must not share projects

that mostly consist of code written by

generative AI tools. Now, obviously, I

think this is a really terrible way to

word anything. What is mostly? Is mostly

51%? Is it 75%?

Do they just mean a plurality? That is

just like the largest portion in

comparison to any other contributor?

It's actually kind of a confusing

phrase. And even more so, what about

projects that were really beautifully

hand artisanally crafted, but now are

being more and more LLM generated with

some light hand edits. I'm thinking of

Ghosty right now. Okay? Uh Mitchell's

made it absolutely clear on the internet

for the last year. Lots of LLM use. So,

at what point does his project go from

being allowed to not being allowed? Is

there a line count? Is there some sort

of specific measurement we can make? Or

do some projects just simply get the old

you're okay cronyism? I don't know. The

second big thing they point out in this

terms of use update is that there's

unclear copyright status. Now, this is

actually honestly the single best

arguments against LLM. See, LLM they can

take code from any type of license and

you know they have, okay? I know a very

specific somebody that would really,

really like to have any code in all code

but yeah.

So, what I mean by that is that there's

things like copy left. Copy left means

that hey, you can use this code but if

you use that code, you also have to be

copy left. It's a viral license. That

means if you inadvertently took code

directly from a copy left repository,

you also would have to make your project

copy left. Now, that can be problematic.

I think everybody can agree that's not

necessarily what you want to see. Now,

with LLM, are you taking code from copy

left repositories?

I don't know. Some countries might have

different opinions on this than other

countries. We're still in quite the gray

area and this seems to be the most like

ethical and sound argument together,

which is we don't know what the license

were to train on it. We don't want to

ruin our project or potentially fall

under some sort of EU jurisdiction or

some Australian law or some law across

the world in which says we can't do that

and then we have to figure out the legal

problems that are in kind of involved in

this conundrum, this quagmire that we

found ourselves in. And so, to me that

actually seems to be the most clear and

consistent possible

active against LLM. So, I actually do

support this. This is a good reason. If

this is Codeberg's complete reason, I

totally get it but it's not actually

their full reason. And furthermore, have

little safeguards to ensure that they do

not include harmful code. Now, this one

obviously I genuinely pretty disagree

with two two reasons. We've seen just a

disasterous amount of CVEs and all sorts

of stuff coming in actually showing just

how vulnerable software really is. We've

seen some of the most epic hacks of all

time including giant tan 75, okay? RIP

giant tan. Now, all of those happened

effectively before LLM. And so, when

people are like, well, there's no

safeguards here, it's just like, well, I

don't know, like how safe is the

software we're using? How safe is the

software you've generated? Generally

speaking, I just have a sneaking

suspicion that software generated is uh

as insecure or slightly more insecure,

but that is my personal bias. But also,

you can just be like, yo LLM, here's a

thousand lines. Can you help find Oh,

but then Fable will tell you can't and

uh OpenAI will also tell you it can't.

Hey, thank you.

Thanks China for letting me review my

own code. Okay. So, when you use the

China models, they will actually tell

you, yeah, hey, there's some problems

potentially right here and here. And

those actually have had a decent success

in finding a lot of stuff. So, using

LLMs can actually be quite beneficial to

ensure the safeguards. So, it really

does kind of put us into this weird I

don't know, it kind of puts us into a

gray zone with this whole terms of use

here. Now, that's just the terms of use.

They also released this blog in which,

of course, like I said, I read it from

left to right, top to bottom, went all

in on it, and they have a couple really

strong points. The first one I think is

losing trust in each other. Effectively,

this statement is all about how building

a community and building trust requires

one-on-one interaction with good

intentions. And a lot of people

submitting say AI-generated stuff, often

well-meaning, put in kind of low effort

and cost a lot of time on the people

doing the reviews. And this kind of

causes a bit of a community upset

because now maintainers want less to do

with people they don't know, which means

that communities will be less welcoming

and people who actually want to try hard

may find it much harder to get involved

in the community due to some accidental

biases that are created in you uh just

because of the very nature of LLMs

themselves and the people who have

wielded them. And so there is this kind

of weird thing that's happening. LLMs in

some sense are breaking down a lot of

the trust barriers and kind of causing

some community fracturing. Codeberg

really cited that as, hey, this sucks.

We don't want that. And honestly, I

agree. That I mean, that's like a really

good reason. They also talk a lot about

like the hardware cost. Codeberg is

completely self-hosted, so the cost of

stuff that used to be $700 for an SSD is

now 3,700 euros for an SSD. I don't know

what a euro is. I'm a freedom unit man,

okay? I like to measure my tea in

gallons, buddy. Now, I don't

I don't know anything about hosting

hardware. I've never done it, but I can

tell you that going from $700 to $3,700

is a gigantic increase, even in

non-freedom units.

And so I can see that the literal cost

of hosting is going up, but then also

they're just being strained all the time

either by people committing in an

outrageous amounts of commits, like if

you've read say the cursor papers or the

Anthropic papers, where they're talking

about thousands of commits per hour,

which is something that say all of

Google used to do. 80,000 engineers are

now being down done by like a handful of

agents. This can cause, obviously, some

real strain on resources. And not only

that, but also the people that are just

constantly crawling the website, they're

also going to cause a lot of resource

problems. And so it's like, yeah,

they're actually like LLMs themselves

are being like a net negative on hosting

code. So this is a genuine problem both

from the user perspective and from the

crawling perspective. Those are two

really good points as to why LLMs are

just kind of destructive in the code

hosting space, both community and cost.

So that totally legit, I can see why

they're saying no to projects for that

very reason. They also highlighted as

for fun this right here, which says that

they're they don't want to use your data

for training. They're not going to use

your data for training. And that is

amazing, because that means it's going

to create a website that is filled to

the brim with hand-crafted, artisanal,

grass-fed, free-range, traditional

coding. This is going to be some of the

highest quality training data known to

man. And I have a sneaking suspicion. I

know exactly who's going to

But there's one part of this blog that I

I severely disagree with. Something that

I I have like a visceral reaction to.

And let me explain why. Together, these

forces make collaboration not only

harder, but also less rewarding. Talking

about the cost, talking about the cost

to maintainers, all of that. With the

transaction cost of collaboration

increasing, people are becoming less

likely to contribute to creating

high-quality software projects and more

likely to vibe code one-off software

that is specific to your need and won't

evolve beyond. We get in a vicious cycle

where collab oration is becoming less

and less rewarding while the amount of

single-use software that is unmaintained

and never sees any improvement is going

up. They say this in kind of a negative

sense and I just fully and completely

and utterly disagree with that

statement. I think the idea of

single-use software is incredible. I

actually think it's fantastic. There's a

bunch of software out there that would

be like demonstrably better if they just

did the thing that it was originally

intended instead of a bunch of people

get in they say, "Well, I want it to do

this. I want to do this. I want to do

this." It's like, "No. Software

sometimes is considered done." This is

done software. Sometimes I want to

create something in which doesn't really

map to any one tool. I don't need to go

and augment a tool to make it fit my use

case. I can just try it out. Does it

feel good? Do I like it? Should I take

this thing seriously? Because let's just

face it, nine out of 10 of my ideas

aren't good ideas and those used to take

me months at a time to really get into a

state where I go, "Oh, yeah. That wasn't

that good." One of those ideas was this

99. I put a lot of effort into creating

this beautiful piece of code in which

goes off and is able to do like these

really kind of strategic LLM calls and

how you interact with the agent and

through Neovim. And I thought that this

could actually be a really enjoyable way

to be able to use LLMs. And lo and

behold, I generally was incorrect on

that. But it was a try. It was an

attempt. I didn't vibe code it though

because I didn't think a lot of the LLM

state of affairs was good enough to vibe

code. I effectively hand-coded the whole

thing. It took me like 2 months to

really try out something and I was

wrong. It hurts a little bit. Looking

back, honestly, probably should have

vibe coded way harder to begin with

until I knew that I liked what I was

seeing. So, I'm actually a huge

proponent of single-use software. I

think single-use software is honestly

incredible. I think we need more

single-use software. I think you should

make all the software you want for

yourself, and then you throw away the

things you don't like, and the things

you really like you start to take

seriously. Those are the things you

should promote into, "Hey, let's put a

little bit of effort into this cuz this

is actually something special." So,

there you go. That is Codeberg's no LLM

in a nutshell. Hopefully, I was able to

actually tell you guys what happened as

opposed to being just some sort of crazy

person on the internet that either says

based or this is the worst thing I've

ever heard. Maybe, just maybe there's a

middle ground somewhere in between, and

maybe a small nonprofit that makes its

money off of donations and a little bit

of membership fees should not be

expected to operate in the same way that

GitHub operates, a multi-billion-dollar

corporation backed by a

multi-trillion-dollar

corporation in which uses that platform

as a means to train their AI to

ultimately attempt to put you out of

business and become a

multi-10-trillion-dollar

company. I'm just saying, maybe we

should have different expectations, and

maybe, just maybe my phone should not go

off in the middle of me going on a

soliloquy here, but also, maybe it's

okay to have different types of

community. Maybe Codeberg will end up

being something special cuz when you go

there, you're going there for the

community, you're going there for the

interaction. You're going there because

you want to become a better software

engineer. And then you can go to other

places to just vibe out something.

Totally fine, brother. The name is I

almost I'm doing a quote from another

kingdom at this point. Again.