Post

What about regular contributors that have llm's do reviews and find problems using tools like GitHub copilot. Does having an AI review a PR count as too much? If so why?

Entirely based IMO

I think this is good for IDK how long, but right now it makes sense. People out there are flooding OSS with sloppy PRs.

Feels like the right line to draw. Assisted research is fine, but the code and the description of what it does should be yours.

I understand it. The health and time of the maintainers is of top importance. That being said, I don't expect this policy to age well. The quality of code and reasoning coming out of LLMs keeps going up. It may not be to JW level yet, but it's surpassed most of us.

Also Fable writing code documentation is horrible using terms like "blast radius", "attack surface" and what not. Am I writing code or making bomb 😬😂

More than just the risk, it's _exhausting_ to have to deal with endless slop PRs.

I'd say "good luck enforcing it"?

There's no way an OSS project could ever review the volume of changes an army of LLM enabled devs could push at them, so quite understand this. Personally I don't read it as an attempt at wider push back against LLMs in development, but I sense many do.

Given how often LLMs are used to diffuse responsibility (including for your own understanding) rather than any practical use, it should be standard. It won't stop people, but if they are busy making sure their LLM output makes sense because it needs to look right, that is a better situation.

It's entirely cost benefit analysis. Are the benefits of the good PRs enabled by AI worth the cost of needing to sift through the slop? Or does it produce better results to just ban LLM generated code so maintainers can direct their finite time and energy elsewhere?