* Jim Porter <[email protected]> [2026-03-10 02:55]: > 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. 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.
While that is the issue, it may not be the real issue at hand. We have to resolve the fundamental problem at its core. Mass of programmers worldwide are stunned by what LLMs generate. And they can't compete with it. And mass of programmers would like to adopt or use technologies, though are lacking proper resources to run it. Those resources could be knowledge, money, hardware and so on. Many things that seem like issues are actually just a lack of knowledge or skill. Once you have the right expertise in a specific area, those problems vanish. ================================================================ The real issue is gaining the knowledge and skills needed to effectively adopt and leverage generative technologies for users' benefit. ================================================================ Back to your problem, where I personally see no problem at all: > One issue here is review burden; if I'm reviewing sloppy > human-generated code, it's usually *very* obvious. 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 solve the review burden traditionally while embracing LLMs, implement automated pre-review tests—such as static analysis, unit test generation, vulnerability scanning, and style compliance checks—directly within the LLM’s output pipeline, so that only vetted, high-quality patches reach the human reviewer, thereby reducing cognitive load and improving both submission quality and review efficiency. In other words -- solve it from both sides: - user having knowledge and skills to use assistive LLM technologies - developer reviewing issues, could guide the user on how to adequately use LLM technologies to submit patches which work Embracing new technologies and not being in fear of it is the key to progress. > More broadly though, I'm concerned that LLM-generated contributions > undermine the social basis of free software. LLM-generated contributions do not inherently undermine the social basis of free software—unless they are deployed without transparency, attribution, or community engagement. The social foundation of free software rests on collaboration, shared understanding, mutual learning, and ethical responsibility—not on the human origin of every line of code. > 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 remember times when I was told that future will come where we instruct computer with natural language as the highest programming language. Maybe it is my hallucination, but I have memories of it. I was waiting long time for that to happen, and there is still so much more to go. In the 1960s and 1970s, projects like SHRDLU (by Terry Winograd, 1972) demonstrated that a computer could understand and act upon natural language commands in a restricted domain, sparking optimism that natural language could one day replace formal programming syntax. https://en.wikipedia.org/wiki/History_of_natural_language_processing So now -- we are in 21st century, the vision from 1960-1970 is not practically here. We shall embrace generative technologies not as a threat to our craft, but as the long-awaited evolution of human-computer collaboration—or risk becoming irrelevant in a world that no longer waits for perfection, but for progress. Jean Louis
