5 ms·

Debian has published the official results for the 2026 GR on LLM usage

14 points by tulio_ribeiro 1mo ago

tulio_ribeiro 1mo ago
Good outcome, mostly. Option E beat literally every other option, so it is an unambiguous Condorcet winner. Options: https://www.debian.org/vote/2026/vote_002 https://www.debian.org/vote/2026/vote_002 The tally.txt file is still unavailable. Bradley–Davidson analysis: https://gist.github.com/tulior/74479b28ddd1e191e5b2d89b8b68ccfd https://gist.github.com/tulior/74479b28ddd1e191e5b2d89b8b68c... The Bradley-Davidson model gives Option 5 a 99.9993% posterior probability of ranking first. Options 2 and 6 are effectively tied for second. Option 5 also beat every alternative head-to-head. Its closest matchup was against Option 2, at 57.8% to 42.2%, so the result was decisive overall even though the closest pairwise contest was not a blowout.
- grahamjperrin 1mo ago
  Is this the tally? https://vote.debian.org/~secretary/gr_llm/results.txt https://vote.debian.org/~secretary/gr_llm/results.txt
  - tulio_ribeiro 1mo ago
    It is. Thanks for the update.
parasxos 1mo ago
For anyone not wanting to parse the ballot page: the winner was Proposal E, "Responsible Use of Generative AI" i.e Debian neither endorses nor prohibits AI tools. Contributors remain fully responsible for anything they submit, and disclosure is encouraged but not required. The outright ban via the Social Contract (Choice 1, which needed a 3:1 supermajority) lost to this decisively.
Conol_ai 1mo ago
[flagged]
Conol_ai 1mo ago
[flagged]
roryirvine 1mo ago
The graph at https://vote.debian.org/~secretary/gr_llm/ https://vote.debian.org/~secretary/gr_llm/ is really useful when understanding the results, too. Debian is remarkable for being such a huge project which has successfully held itself together for so long. GRs like this are an essential part of that, allowing controversies to be settled in a way that manages to be both fair and transparent.
NitpickLawyer 1mo ago
> That supports “decisive win” comfortably; whether it qualifies as a “landslide” depends on where the margin threshold is set. I think the landslide is "pro" vs. "against". The two most "against" options got the lowest rank. Top 4 are all "pro" (with different caveats for each). Rank Option θ i 1 5: Responsible Use of Generative AI 1.794 2 2: Allow AI-Assisted Contributions with conditions 1.430 3 6: A cautious approach to generative AI 1.424 4 4: Accept AI contributions for Debian specific work 1.252 5 8: Avoid LLM use: climate impact 1.007 6 7: Debian is created by humans 0.881 7 9: None of the above 0.762 8 3: Reject LLMs as far as practical 0.683 9 1: Ban LLM contributions via Social Contract 0.474
- tulio_ribeiro 1mo ago
  Great insight. Let me see what ChatGPT can cook if we classify each. The tally is now available; might be useful. ### Edit Re-classified using permit, prohibit, materially_restrict or none_of_the_above as I think these are more accurate labels. Conclusion: All five proposals that continued to permit AI-assisted direct contributions occupy ranks 1 through 5. The result is: permit: 15 wins restrict/prohibit: 0 wins Every single permit option beat every single restrictive or prohibitive option head-to-head. ChatGPT’s interpretation: Debian was divided over how AI-assisted contributions should be handled, but not over whether they should remain permissible: all five permissive proposals outranked every proposal that materially restricted or prohibited them, and permissive proposals won all 15 cross-category head-to-head comparisons.

14 points by tulio_ribeiro 1mo ago

Good outcome, mostly. Option E beat literally every other option, so it is an unambiguous Condorcet winner. Options: https://www.debian.org/vote/2026/vote_002 https://www.debian.org/vote/2026/vote_002 The tally.txt file is still unavailable. Bradley–Davidson analysis: https://gist.github.com/tulior/74479b28ddd1e191e5b2d89b8b68ccfd https://gist.github.com/tulior/74479b28ddd1e191e5b2d89b8b68c... The Bradley-Davidson model gives Option 5 a 99.9993% posterior probability of ranking first. Options 2 and 6 are effectively tied for second. Option 5 also beat every alternative head-to-head. Its closest matchup was against Option 2, at 57.8% to 42.2%, so the result was decisive overall even though the closest pairwise contest was not a blowout.

grahamjperrin 1mo ago
Is this the tally? https://vote.debian.org/~secretary/gr_llm/results.txt https://vote.debian.org/~secretary/gr_llm/results.txt
- tulio_ribeiro 1mo ago
  It is. Thanks for the update.

parasxos 1mo ago

For anyone not wanting to parse the ballot page: the winner was Proposal E, "Responsible Use of Generative AI" i.e Debian neither endorses nor prohibits AI tools. Contributors remain fully responsible for anything they submit, and disclosure is encouraged but not required. The outright ban via the Social Contract (Choice 1, which needed a 3:1 supermajority) lost to this decisively.

Conol_ai 1mo ago

[flagged]

Conol_ai 1mo ago

[flagged]

roryirvine 1mo ago

The graph at https://vote.debian.org/~secretary/gr_llm/ https://vote.debian.org/~secretary/gr_llm/ is really useful when understanding the results, too. Debian is remarkable for being such a huge project which has successfully held itself together for so long. GRs like this are an essential part of that, allowing controversies to be settled in a way that manages to be both fair and transparent.

NitpickLawyer 1mo ago

> That supports “decisive win” comfortably; whether it qualifies as a “landslide” depends on where the margin threshold is set. I think the landslide is "pro" vs. "against". The two most "against" options got the lowest rank. Top 4 are all "pro" (with different caveats for each). Rank Option θ i 1 5: Responsible Use of Generative AI 1.794 2 2: Allow AI-Assisted Contributions with conditions 1.430 3 6: A cautious approach to generative AI 1.424 4 4: Accept AI contributions for Debian specific work 1.252 5 8: Avoid LLM use: climate impact 1.007 6 7: Debian is created by humans 0.881 7 9: None of the above 0.762 8 3: Reject LLMs as far as practical 0.683 9 1: Ban LLM contributions via Social Contract 0.474

tulio_ribeiro 1mo ago
Great insight. Let me see what ChatGPT can cook if we classify each. The tally is now available; might be useful. ### Edit Re-classified using permit, prohibit, materially_restrict or none_of_the_above as I think these are more accurate labels. Conclusion: All five proposals that continued to permit AI-assisted direct contributions occupy ranks 1 through 5. The result is: permit: 15 wins restrict/prohibit: 0 wins Every single permit option beat every single restrictive or prohibitive option head-to-head. ChatGPT’s interpretation: Debian was divided over how AI-assisted contributions should be handled, but not over whether they should remain permissible: all five permissive proposals outranked every proposal that materially restricted or prohibited them, and permissive proposals won all 15 cross-category head-to-head comparisons.