Remix.run Logo
▲ linsomniac 4 hours ago

>My open source projects were barely sustainable

AI can help with that! Give an agent some rules, like via a skill for the project, and then point it at the PR/Issue for triage to determine if it's one that fits the rules for auto-closing (already reported and decided WONTFIX, for example). Have it run this across all your repos.

Have the AI tooling do the heavy lifting of reproduction and benchmarking. One of my projects is fairly complicated and one user ran into some problems which he could describe the situation that caused it, but was light on any other details: my apt package cache was slow to serve the PHP update to his users. I asked AI to try to reproduce it: Set up a repo, set up different versions of my cache, have a client pull it, put an updated package in there, and have a client pull it again under different configurations. This could have taken me hours to do myself. AI did it with ~2 minutes of my effort.

>I have not organized my life around triaging annoyances

Then don't. AI can help with the triaging. Or, I give you permission: Close the repo for contributions. If you don't have time to maintain it, let people know, so they can fork or replace it. It sounds like your repos are already closed to contributions, due to lack of time/attention. That's fine, don't feel bad, just go ahead and make it official.

▲flatline 4 hours ago | parent | next [-]

You can say, "go off and fork the project", and you'll get 20 forks, each one person's pet PR applied, and no more future progress without a dedicated maintainer triaging and coordinating all this stuff in one place. You are asking projects that are surviving but not necessarily thriving to simply die. And for what? To satisfy downstream developer demand for their AI-generated contributions to be accepted?

Not all contributions are equally valuable, and that is notwithstanding AI use. AI facilitates volume but not necessarily quality. Why should anyone spend resources benchmarking and reproducing a non-issue, or a PR that will not otherwise pass muster? We can automate a lot, but human judgment is still critical in making things to be used by humans.

▲linsomniac 2 hours ago | parent | next [-]

>you'll get 20 forks, each one person's pet PR applied

Sure; if the primary repo goes unmaintained, and nobody else steps up to maintain a fork, yes you might get multiple forks that solve a specific users individual issue. I'm not sure there's a way around that.

The comment I was replying to sounded like someone who was unavailable to handle non-AI requests as well. AI contributions is making it worse, though my biggest projects haven't been hit by it yet <knock on wood>.

Low quality contributions is a very real problem, I agree with that.

▲zem 2 hours ago | parent | prev | next [-]

how does the presence of those forks contribute to the project's demise in any way?

▲hardbass an hour ago | parent [-]

It was common point of irritation and frustration to see projects like GNOME and Mozilla make bs excuses for not fixing bugs and randomly closing bug requests. People writing their own forks and patches were hailed as doing a worthy act. After AI, suddenly those people are all evil and the GNOME and Mozilla attitude is correct.

▲hardbass 3 hours ago | parent | prev [-]

Of course. So do you want a Grand Vizier to ban forks that are not valid so software runs the way it should be run?

▲yeoldeprogramme 4 hours ago | parent | prev | next [-]

“ he could describe the situation that caused it, but was light on any other details”

You are too nice. Anyone who isn’t going to spend time to explain an issue should be automatically rejected.

We’re in this ugly situation where AI causes the problem and people’s idea to fix it is throw in more AI.

Just like throwing technology at climate change when our own obvious habits are destroying the planet. Let’s move to Mars.

▲necovek 4 hours ago | parent | prev | next [-]

By running automated AI agents on untrusted inputs (PRs from strangers on a public repo), you open yourself up to attacks. And you could be spending money for somebody else.

▲linsomniac 2 hours ago | parent | next [-]

Agreed and this is a very real thing to be concerned about.

I like the idea of leaning on a decision like Jev model to identify such things.

▲epistasis 3 hours ago | parent | prev [-]

The only attacks possible would be the tools given to the agent.

▲swiftcoder 2 hours ago | parent [-]

Assuming it can't perform a sandbox escape, at any rate. Maybe a Jev-like decision model would be a better idea.

▲epistasis an hour ago | parent [-]

Or just don't allow tools, write a simple python script that uses the API and doesn't do tool calls, etc. etc. etc.

Table stakes for being an engineer these days should include knowing how to control LLMs, it's as fundamental as a working knowledge of databases. One doesn't need to know how design a data model from scratch or code up that python script using the OpenAI API from memory, but should know the general shape of what's possible and how to look it up.

▲matkoniecz 4 hours ago | parent | prev | next [-]

> AI can help with that! Give an agent some rules, like via a skill for the project, and then point it at the PR/Issue for triage to determine if it's one that fits the rules for auto-closing (already reported and decided WONTFIX, for example). Have it run this across all your repos.

For me it was a bit useful but it has not offset even 10% of damage caused by LLMs

and is solving least important problem

▲aleph_minus_one 3 hours ago | parent | prev [-]

> AI can help with that! Give an agent some rules, like via a skill for the project, and then point it at the PR/Issue for triage to determine if it's one that fits the rules for auto-closing (already reported and decided WONTFIX, for example). Have it run this across all your repos.

This requires a huge setup of the necessary AI infrastructure.