Policy on vibe coded nix tools?

The announcement from authors that software is vibe-coded or ai-assisted is already, itself, a categorization, self imposed by the author. So it’s just begging for a forum category really.

Code produced purely with AI is either difficult to maintain, or won’t be maintained. Similar to proprietary software or blobs, which already has its own category in meta.unfree. So there should be a category for posts pertaining to proprietary software, and-or vibe-coded software.

Why not contain it to a category as Haiku have? It seems to be having a positive impact on their discourse.

New Category for "AI-Assisted" Software? - Websites - Haiku Community

8 Likes

The dirty secret of our industry is that, asymptotically, all code falls into one of those two buckets.

3 Likes

Fair, though I do think it is less selfish and more responsible to produce code by your own hands and fail to maintain it. Often, attention is being demanded for a result that took no effort to create, for which the author does not imply any commitment whatsoever; at the cost of a large amount of electricity that they didn’t pay for. It is also a result that attempts to further steal attention from users who may be unable to judge.

The effort that has gone into creating something, is part of moral judgement.

And I do think it’s valid to judge. People can be judged based on their character, and their mindfulness. Using AI to generate unmaintainable code, advertise unethical products and services, (or to contribute incorrect information to wikipedia for example) has become part of the character of maintainers, and a sign of low mindfulness or just plain irresponsibility in some cases.

In the best case, it’s like someone obsessing over and advertising what text editor they use, and I wouldn’t enjoy that either.

6 Likes

I also wonder whether it’s compatible with some of the values outlined in Governance | Nix & NixOS

1 Like

now, one of them is currently phrased as:

Automation over process and toil

whether that was written with our current situation in mind (or what current popular support on specific values is like), i can’t tell.

which ones did you have in mind?

1 Like

In mind, I had:

Stable evolution over stagnation or chaos”.

Containing AI generated content to a category has had a positive impact on the Haiku discourse from my perspective. The decision, that some protested at first, has been completely fine, and the result has been that the people who don’t want to see it, can filter it, and the people who want to see it, are not affected.

Not doing this, just leads to more chaos, in my opinion, and the longer it is left unaddressed, the more outcry there will be at the decision ultimately being made.

The constant demand by people for this to be categorized, and the lack of any action, is chaotic.

Another one I had in mind was:

Individuals gain trust and status by doing the work

Letting tools that you’re not in control of and are not even self-hosting, do the work, but demanding the same attention and respect is actively eroding trust.

Nixpkgs contributors are a large community of real people who have had their contributions taken by this tool. This tool is then generating regurgitated, derivative work that some are demanding attention be given to, and there is also an expectation of positive feedback, which doesn’t seem right to me. It seems like showing off to the people you stole from, which is a bit rude.

If real developers, who are not using these tools, stop writing and publishing quality code examples for training, then quality output also drops or stops.

6 Likes

It seems like showing off to the people you stole from

I think this raises the question on why we code. Gaining status, as you mentioned, could be one motivation. Another though might be offering tools to empower others. In such a reading, one could potentially take less offense than asserted here, and regard the process as less of a competition.

To be fair, I don’t necessarily have data on where others might stand there.

2 Likes

Well, most of the time, I want to empower me. And as I already have written the code to do so, I publish it, such that others do not have to repeat the work.

Empowering others is therefore just a sideeffect of empowering me in the first place.

1 Like
8 Likes

let me get this straight… all LLM-assisted code is AI slop, unless it’s my LLM-assisted code since clearly I’m eminently qualified as a contributor? garbage open source contributions have existed since the beginning of time, no AI required. and during that time we’ve never required “hey I’m a little bit stupid” disclosures in a PR description.

the increase in volume to filter through is certainly real, but nonetheless AI-assistance seems like the wrong signal to isolate here. at the end of the day, solid contributors will produce solid output regardless of their path to get there, and I’m fairly oppose d to anything that results in those contributors deciding that it’s not worth their time to contribute “manually” vs. alternative ways to spend their time.

I’m quite fond of @mitchellh’s approach to this one with GitHub - mitchellh/vouch: A community trust management system based on explicit vouches to participate. · GitHub – the set of relevant contributors hasn’t substantially changed, we just need to select them more carefully than before given the volume.

3 Likes

Humans and LLMs can create shitty code the same, indeed.

Though having used a LLM can have legal implication that have yet to be explored. Like, as far as I understand, the code generated by an LLM is not “owned” by anyone according to US courts, but still whoever uses it, if it is found that the snippet is violating some license, you can get made responsible.

Therefore I understand the demand of disclosure.

4 Likes

I agree, and it can have moral implications too. It’s also a good indicator of whether something’s worth investing your time in.
I do think AI code tends to worse quality than human code, though I acknowledge a talented human can utilize AI to produce high quality code. But I also don’t care. Quality is one drop in the bucket of reasons I don’t like AI.

1 Like