Add an AI policy #3

MultisampledNight commented 2026年06月20日 15:21:03 +02:00

• Make sure no generated code ends up in repo or in nonquarantined communications. Contributions with generated code/docs/comms will be treated like a formal problem as per SFC LLM-gen-AI recommendation 3, i.e. responded to with kindness and rejected
• Inform users of the harm they are perpetuating by using LLMs (possibly make them watch https://www.youtube.com/watch?v=olhu9UhFGl4)
  • Overview: https://blog.johanneslink.net/2025/11/04/to-gen-or-not-to-gen/
  • Human rights violations: https://www.amnesty.org/en/documents/pol40/0996/2026/en/
  • Economic harm: https://www.mcsweeneys.net/articles/ai-economics-for-dummies, https://www.wheresyoured.at/the-case-against-generative-ai/
  • Social harm: https://nicolasbouliane.com/blog/death-by-ai, https://joshcollinsworth.com/blog/sloptimism, https://www.garfieldtech.com/blog/selfish-ai
    • Most software is safety critical: https://www.softwaremaxims.com/blog/we-should-have-caught-it
  • Epistemic harm: https://ergosphere.blog/posts/the-machines-are-fine/, https://dl.acm.org/doi/10.1145/3442188.3445922
• If despite this, they want to use them anyway, force the LLM to act in a TA-only mode. Make the user write all code (other than bare symbol autocompletion) and puncture their understanding at times (see also https://htmx.org/essays/universities-and-ai/)
• If a user can give a convincing argument as for how using an LLM generatively reduces a disability barrier for them, have them get in touch with the repository maintainer via social private mention or email.

Inspiration:

• Language to pick
  • imagine you're talking to a neoliberal: https://unstable.systems/@AmyZenunim/116672510693285709
  • use pseudo-reasoning shortcuts: https://needtoknow.fyi/card/shortcut-reasoning/
• What Carson Gross uses; probably still too lax: https://gist.github.com/1cg/a6c6f2276a1fe5ee172282580a44a7ac
• https://github.com/DrCatHicks/learning-opportunities

- *Make sure **no** generated code ends up in repo or in nonquarantined communications.* Contributions with generated code/docs/comms will be treated like a formal problem as per [SFC LLM-gen-AI recommendation 3](https://sfconservancy.org/llm-gen-ai/llm-backed-generative-ai-recommendations.html), i.e. responded to with kindness and rejected - Inform users of the harm they are perpetuating by using LLMs (possibly make them watch https://www.youtube.com/watch?v=olhu9UhFGl4) - Overview: https://blog.johanneslink.net/2025/11/04/to-gen-or-not-to-gen/ - Human rights violations: https://www.amnesty.org/en/documents/pol40/0996/2026/en/ - Economic harm: https://www.mcsweeneys.net/articles/ai-economics-for-dummies, https://www.wheresyoured.at/the-case-against-generative-ai/ - Social harm: https://nicolasbouliane.com/blog/death-by-ai, https://joshcollinsworth.com/blog/sloptimism, https://www.garfieldtech.com/blog/selfish-ai - Most software is safety critical: https://www.softwaremaxims.com/blog/we-should-have-caught-it - Epistemic harm: https://ergosphere.blog/posts/the-machines-are-fine/, https://dl.acm.org/doi/10.1145/3442188.3445922 - If despite this, they want to use them anyway, force the LLM to act in a TA-only mode. Make the user write *all* code (other than bare symbol autocompletion) and puncture their understanding at times (see also https://htmx.org/essays/universities-and-ai/) - If a user can give a convincing argument as for how using an LLM generatively reduces a disability barrier for them, have them get in touch with the repository maintainer via social private mention or email. Inspiration: - Language to pick - imagine you're talking to a neoliberal: https://unstable.systems/@AmyZenunim/116672510693285709 - use pseudo-reasoning shortcuts: https://needtoknow.fyi/card/shortcut-reasoning/ - What Carson Gross uses; probably still too lax: https://gist.github.com/1cg/a6c6f2276a1fe5ee172282580a44a7ac - https://github.com/DrCatHicks/learning-opportunities

MultisampledNight commented 2026年06月20日 15:30:23 +02:00

I don't want to ban AI overall. We, as a generation of antifascists, mustn't fall into a "generational blunder": https://www.youtube.com/watch?v=lRq0pESKJgg

I am not a friend to the technofascists. I am exactly the demographic that they are trying to make disposable. But I'm afraid that if we aren't careful with how we critique the use and abuse of artificial intelligence, we might end up, as leftist philosopher Mark Fischer once warned, foreclosing the possibility of a technologized anti-capitalism. I fear that framing this debate as anti-tech versus pro and handing the pro tech position to the right-wing is a generational blunder. (5:31)

Hating AI for the wrong reasons isn't just a philosophical blunder, but the first step down a path that may end up hurting the very people we vowed to protect. (1:12:25)

I do think that using AI for narrow tasks is useful. I also think that an introspective instead of generative approach could be genuinely valuable for swdev (I think the point of writing FOSS is to collectively learn to write better code and systems, not to actually have a product, hence I also think that categorically denying generative use is defensible).

I don't want to ban AI overall. We, as a generation of antifascists, mustn't fall into a "generational blunder": https://www.youtube.com/watch?v=lRq0pESKJgg > I am not a friend to the technofascists. I am exactly the demographic that they are trying to make disposable. But I'm afraid that if we aren't careful with how we critique the use and abuse of artificial intelligence, we might end up, as leftist philosopher Mark Fischer once warned, foreclosing the possibility of a technologized anti-capitalism. I fear that framing this debate as anti-tech versus pro and handing the pro tech position to the right-wing is a generational blunder. (5:31) > Hating AI for the wrong reasons isn't just a philosophical blunder, but the first step down a path that may end up hurting the very people we vowed to protect. (1:12:25) I do think that using AI for narrow tasks is useful. I also think that an introspective instead of generative approach could be genuinely valuable for swdev (I think the point of writing FOSS is to collectively *learn* to write better code and systems, not to actually have a product, hence I also think that categorically denying generative use is defensible).

MultisampledNight commented 2026年07月01日 21:35:36 +02:00

How other projects are doing it:

• Godot: https://godotengine.org/article/contribution-policy-2026/
• See https://github.com/rust-lang/rust-forge/pull/1040 description section Prior art

How other projects are doing it: - Godot: https://godotengine.org/article/contribution-policy-2026/ - See https://github.com/rust-lang/rust-forge/pull/1040 description section `Prior art`

MultisampledNight commented 2026年07月03日 03:37:33 +02:00

After reading https://doi.org/10.1007/s00146-023-01710-4: This issue and #2 are the exact same, just different perspectives on "what values do we follow in this project". We don't need to look for primordial argumentation on why or why not LLMs could ever develop, because that is irrelevant at this stage: We care about ethics and politics, first and foremost, and metaphysics somewhere after that. We can absolutely construct values to use as root for metaphysics

After reading https://doi.org/10.1007/s00146-023-01710-4: This issue and #2 are the exact same, just different perspectives on "what values do we follow in this project". We don't need to look for primordial argumentation on why or why not LLMs could ever develop, because that is irrelevant at this stage: We care about ethics and politics, *first and foremost*, and metaphysics somewhere *after* that. We can *absolutely* construct values to use as root for metaphysics

MultisampledNight commented 2026年07月03日 03:37:47 +02:00

Hence closing in favor of #2

Hence closing in favor of #2