▶ YouTube Summaries

• YT :: https://www.youtube.com/watch?v=h3DlwC1nex0
• Original title :: The Codeberg Situation | TheStandup

Prime, Casey Muratori, Trash Dev, and TJ (Teach) dig into Codeberg's new addendum banning projects that mostly consist of generative-AI code. They walk through Codeberg's stated reasons (hardware and infrastructure strain, community erosion, unclear copyright, single-use software), largely find the reasoning sound even if unenforceable, and spend most of the conversation on the harder question underneath: why humans keep programming by hand even when a machine could do it faster, and why that choice deserves no apology.

The addendum and the vote

Codeberg's new rule: no sharing of projects that mostly consist of generative-AI code, citing unclear copyright status and insufficient safeguards against harmful code. The controversial line triggered a pile-on online. A poll image showing roughly two-thirds support was quietly removed from the blog post afterward; Casey notes the two- thirds figure was of those who voted, roughly half of eligible members, so more like two-thirds of half.

Codeberg's stated reasoning

The blog post ("Protecting our FLOSS commons from LLMs") gives several reasons: hardware costs are spiking (SSDs going from 700 to 3500 euros) while LLM-driven commit volume rises; a single AI-assisted project can balloon to hundreds of thousands of lines maintained by one person, consuming resources disproportionate to the community value it creates, unlike projects like Ghostty with many contributors; copyright status of AI output is unclear; and they don't want to host single-use, throwaway software built for one person's narrow need instead of shared, extensible tools.

The point the group calls most interesting: loss of trust between maintainers and contributors. Getting burned by low-effort, AI- generated pull requests makes maintainers suspicious of all new contributors by default, which quietly poisons community dynamics — even well-written PRs now get read as "probably slop."

Can this even be enforced

Everyone agrees the reasoning is sound but the "mostly AI" line is unenforceable — nobody can prove how a contribution was made, and the policy risks screening out newcomers who lack a reputation while letting existing, trusted maintainers who openly use AI (like Ghostty, whose creator Mitchell Hashimoto has talked for over a year about using AI to generate patches) continue unquestioned. Casey suggests a simpler alternative goal: nobody should be able to tell a contribution was AI-assisted — meaning a human curated it into a clean, non-verbose, mistake-free submission. Anything past that baseline (repo-house-rules, discouraging AI in general) is Codeberg choosing to invest in cultivating human contributors rather than optimizing for throughput, similar to Zig's stated policy of investing in people who will become long-term contributors rather than accepting AI-generated patches.

They also flag an asymmetry in reaction: everyone assumes their own AI-assisted PRs are fine and everyone else's are lazy slop — the same bias people show toward driving, drinking, or gun ownership ("I do it responsibly, but watch out for everyone else").

Getting "left behind"

The internet's other big reaction: Codeberg will get "left behind" if it excludes AI-heavy contributors, because that's supposedly how everyone will code going forward. The group pushes back hard on this framing, calling it a rerun of Silicon Valley absolutism (e.g., "if you don't use React, no one will work on your project"). Casey argues getting left behind only applies if your goal is VC-style hypergrowth; if your goal is a sustainable place for humans to share code with other humans, you can hit that goal without maximizing headcount. He also makes a broader argument for diversity of practice: humanity benefits from some people opting out of the dominant trend by default, because it hedges against the dominant trend turning out to be a catastrophic mistake nobody could foresee — citing the Amish choosing not to use electricity as a working example, not a punchline.

Self-consciousness, not self-interest, drives the backlash

Casey argues that AI enthusiasts arguing against people who choose not to use AI are acting against their own interest: if you love AI, you should want people preserving non-AI skills and training data, the same way people who love Unity should be glad someone built game engines from scratch instead of discouraging them. He reads the "you must adopt this or be left behind" rhetoric as projection — people who haven't made peace with their own AI use need everyone else to agree it's fine too. He quotes George Hotz's essay on the same theme: hype about a closing window or a "perpetual underclass" is manufactured negative pressure, not a real technical claim.

Circle-of-expertise blindness

Prime frames the disconnect through a "gentleman amnesia" analogy: people rate a magazine article as brilliant when it covers a topic they don't know and shallow when it covers one they do, then forget that pattern applies to themselves. People whose own work maps well onto LLM strengths (CRUD, glue code) assume everyone's work does, and can't imagine why a maintainer wouldn't automate PR review the same way they did.

Why hand-write code at all

The closing thread, credited partly to TJ: programming by hand is like the Amish still making chairs, or people learning to play drums instead of using a drum machine, or playing chess despite computers being better at it — the point was never to be the most efficient possible route to the output. Casey says this directly: he wants to program, not just get working software, the same way playing a losing game of chess against a stronger opponent doesn't make the game pointless. Wanting to do something yourself, even when a machine could do it faster or better, needs no justification — and people who choose that path aren't blocking anyone else's road to a fully-AI-generated future.