* Lars Noodén via libreplanet-discuss <libreplanet-discuss@libreplanet.org> [2025-07-10 21:39]: > On 7/7/25 04:17, Akira Urushibata wrote: > > Maybe I'm not looking hard enough. If anybody can point out to good > > examples of AI usage in free software development, I'd like to know. > > Hi, Akira, > > Yes, LLM-based output has been affecting free software development for a > while. Daniel Stenberg of cURL fame has written about its impact several > times with clear examples. Here are two posts about it: > > "The I in LLM stands for intelligence" > https://daniel.haxx.se/blog/2024/01/02/the-i-in-llm-stands-for-intelligence/ > > "Curl takes action against time-wasting AI bug reports" > https://www.theregister.com/2025/05/07/curl_ai_bug_reports/
Surely, I acknowledge the report from Curl team. Though let us not generalize, so I cannot confirm your statement "LLM-based output has been affecting free software development for a while". Why generalize? It is not any output "affecting" Curl team in their development, but irresponsible reporters. That is isolated case, there is absolutely no need for this generalization. It is false! Not proven. Text generating systems can generate garbage, that depends of human, not the LLM, so blame should be on uneducated LLM operators, not the LLM itself. > Now if you are looking for /positive/ examples the effects of AI slop in > free software projects, I think your hunch is correct and there may not be > any. I was thinking of using this word: The noun gibberish has 1 sense (no senses from tagged texts) 1. gibberish, gibber -- (unintelligible talking) But instead, let me give you some straighforward examples disapproving your statement how there may not be any positive examples of new technologies like Large Language Model (LLM) and akin: [2410.02091] The Impact of Generative AI on Collaborative Open-Source Software Development: Evidence from GitHub Copilot https://arxiv.org/abs/2410.02091 The Impact of Generative AI on Open-Source Community Engagement <br> by Karthik Babu Nattamai Kannan, Narayan Ramasubbu :: SSRN https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5106441 New Study Shows Open Source AI Is Catalyst for Economic Growth https://about.fb.com/news/2025/05/new-study-shows-open-source-ai-catalyst-economic-growth/ Open Source AI: Democratizing the Future of Artificial Intelligence | by Kiplangat Korir | Medium https://medium.com/@kiplangatkorir/open-source-ai-democratizing-the-future-of-artificial-intelligence-7deaa45bd6da Impact of Open-Source AI on Communities https://www.arsturn.com/blog/analyzing-the-community-impact-of-open-source-ai-projects Benefits and Applications of Open Source AI | Moesif Blog https://www.moesif.com/blog/technical/api-development/Open-Source-AI/ > Going back to your question about the beast in Redmond, which although > proprietary, it shows that LLMs get in the way of development, at best. One > can conclude that because competition there between departments and even > employees is so cutthroat that if there had been even the slightest > advantage from using LLM slop oneself, then it would be found all over the > place. It is found all over the place. You have got very subjective opinion, fine, you may have, but hey, does it really matter? Studies have shown that LLM-assisted developers often write code faster, with fewer errors, and spend less time on boilerplate. LLMs are widely used in development tasks like code generation, documentation, refactoring, and testing. However, they can get in the way if: - Used improperly (e.g., blindly trusting outputs). - Replacing deep thinking or careful architecture. LLMs can help or hinder, depending on usage. So your blanket claim is subjective and overgeneralized. > "Because competition... is so cutthroat... if there had been even the > slightest advantage... then it would be found all over the place." This argument is speculative and based on assumptions: Yes, Microsoft (the "beast in Redmond") is known for a competitive culture. I get where you're coming from, but I’ve got to push back — I run an LLM locally on my own machine, completely offline, and it’s been a game-changer for me. The idea that LLMs "get in the way" just doesn’t hold up in practice. And let’s not forget, even Microsoft — the so-called beast in Redmond — has been actively contributing to the field with solid free models like Phi, which are compact, high-performing, and openly available. Just because you don’t see widespread internal adoption doesn’t mean they’re not being used or useful; plenty goes on behind the scenes, and the value is there if you actually try it. Reference: microsoft/phi-4 · Hugging Face: https://huggingface.co/microsoft/phi-4 > However, because LLM slop output does not help, you now see the > creation of mandates there for the remaining employees to force them > to try to (or pretend to) use LLM slop: > > "Microsoft Makes AI Usage Mandatory for Employees as Performance Reviews > Face Overhaul" > > https://medium.com/@hamza_83953/microsoft-makes-ai-usage-mandatory-for-employees-as-performance-reviews-face-overhaul-8f8cc020f637 And? That is their policy, given for you to freely discuss it. They got reasons for it. If you work there and you don't like it, feel free to resign. It is private business. > I've only ever done limited scripting myself, albeit over a longer > period. Thanks for sharing that — but if I may be honest, how can someone who's only done limited scripting, even over a longer period, draw such broad conclusions with confidence? That doesn’t exactly make you an expert, and it’s important to acknowledge the limits of experience when making sweeping claims. > Over that time I gather that while typing in working code is a > challenge, understanding and defining the problem(s) to be solved > and then breaking them down into small units is much harder and so > is documenting the code as you go along. Bingo — you nailed it with that one. Defining the problem clearly, breaking it down into manageable parts, and documenting it properly is way harder (and more important) than just typing out working code. That part, I fully agree with — well said. > Those two aspects are nothing which plausible sentence generators > and plagiarism engines can help with, even if their > attribution-stripped output does seem to compile. I’d disagree there — from my own hands-on experience running an LLM locally, it’s *precisely* those earlier stages — breaking down problems, brainstorming approaches, even drafting clean documentation — where the tool shines most. It’s not just about generating code that compiles; it’s a thinking companion that helps structure ideas, catch oversights, and speed up problem-solving. Calling it a "plausible sentence generator" really undersells what these models can do when used well. > Currently LLM coding is a scam which is being perpetrated because so > few understand ICT and, because of the influence of Redmond against > education, that number is declining. Ah, good to know you've uncovered the grand LLM coding scam — all that with only limited scripting experience under your belt! Impressive detective work. I guess those of us actually using LLMs productively must have missed the memo from Redmond about dismantling education while secretly boosting our workflows. > Software freedom has been slowing that decline but Software Freedom > promoting groups need to team up with educational projects and eject > the productivity killing and knowledge killing microsofters or at > least keep them at bay so as to turn that around and /increase/ the > numbers of those who understand and can use ICT. Right — because nothing says “promoting education and freedom” like gatekeeping who gets to contribute. Good thing you’re here to save the day from the productivity-killing hordes, even though some of us are just quietly getting things done — using whatever tools actually work, including those made by the so-called “microsofters.” Personally, I’ve disliked Bill Gates for about 25 years now — mainly because of the shady tactics he used to monopolize computing and crush competition back in the day. But hey, credit where it’s due: Microsoft today is putting out solid contributions like the Phi models and actually supporting the open-source ecosystem in a meaningful way. Just check https://huggingface.co/microsoft — it’s a prime example of how far things have shifted. Jean Louis _______________________________________________ libreplanet-discuss mailing list libreplanet-discuss@libreplanet.org https://lists.libreplanet.org/mailman/listinfo/libreplanet-discuss
