On 3/7/2026 11:45 AM, Ihor Radchenko wrote:
I do not see this being a problem with LLMs. If someone is pushing
changes carelessly, that's not acceptable. With or without LLMs.
One issue here is review burden; if I'm reviewing sloppy human-generated
code, it's usually *very* obvious. Even so, I can often still see the
intent behind it, so it's worth some extra effort to guide an
inexperienced contributor toward writing an acceptable patch. With
LLM-generated code, the patch is often cleaner (at least superficially),
which in my experience requires much closer attention from the reviewer.
To some degree, this is unavoidable. People may post LLM-generated
patches even if we tell them not to. However, in projects where
LLM-generated patches are banned, reviewers are more able to reject the
patch and move on, rather than expending the extra energy to suss out
all the lurking bugs.
I get it that some people may be overconfident with LLMs, but that's
simply a sign of limited experience. Once you work with LLMs long
enough, it becomes very clear that blindly trusting the generated code
is a very, very poor idea.
I fear this is too optimistic. Projects like OpenClaw and Gas Town were
made by experienced developers who, by now, have certainly used LLMs
extensively. Despite that, the authors rarely (if ever) even look at the
generated code. While not everyone goes that deep down the rabbit hole,
evidently some developers find it irresistible.
---
More broadly though, I'm concerned that LLM-generated contributions
undermine the social basis of free software. The first essential freedom
is "The freedom to study how the program works, and change it so it does
your computing as you wish." In an individualistic sense, LLMs don't
hurt this, and could even be seen as helping; now, non-programmers can
change a program more easily.
However, in a social sense, I believe that we (maintainers,
contributors, etc) have a responsibility to help cultivate these
freedoms in others. The very reason we can evaluate the merit of
LLM-generated code is because we've had the time to hone our skills.
Those who haven't had those years of practice deserve our attention and
guidance so that they too can have those skills. So that they've
developed the *positive* freedom to study and change how a program
works. I don't think there's any better way to do this than to learn by
doing alongside the experts.
I'd much rather my limited time and energy go towards building up the
next generation of free software hackers than to reviewing the output of
a statistical model so I can root out all the highly-plausible but
nevertheless incorrect bits.
(I've tried to keep this somewhat brief, so I hope the above doesn't
omit some essential part of my reasoning.)
