Post B9vJuXtDUFSrys6U6a by alextecplayz@techhub.social
 (DIR) More posts by alextecplayz@techhub.social
 (DIR) Post #B9vFfuzwt6mBUT1RS4 by 9to5linux@floss.social
       0 likes, 0 repeats
       
       #Debian Project Formalizes Responsible Use of Generative AI https://9to5linux.com/debian-project-formalizes-responsible-use-of-generative-ai#OpenSource #Linux
       
 (DIR) Post #B9vJuXtDUFSrys6U6a by alextecplayz@techhub.social
       0 likes, 0 repeats
       
       @9to5linux "Every submission, AI-assisted or not, is still expected to meet Debian‘s usual standards for quality, correctness, maintainability, and legal compliance, and using an AI tool doesn’t reduce the contributor’s responsibility for what they submit.According to the general resolution published today by the Debian Project, blindly uploading AI-generated material without review is explicitly called out as inconsistent with how Debian already operates. Disclosure of AI assistance is encouraged, not required."Which means that no, someone trying to upload a package to Debian that's just a bunch of commits from Claude (e.g. ship fast and break things) likely won't make it into Debian because it's just garbage slop. Disclosure would have been nice to be required, though."The text also sidesteps the unresolved legal questions around AI output, copyright, and training data, leaving contributors to use their own judgment on provenance and licensing, with Debian’s existing policies on licensing and software freedom still applying regardless of tooling."I'd rather trust lawyers and the judicial system to handle the legal part of AI, and whether it's infringing copyright, using stolen or unlicensed training data, etc.And there are 'open' LLM models such as IBM Granite which according to Stanford's Foundation Model Transparency Index, IBM scores the highest of all tested companies in 2025, at 95% (https://crfm.stanford.edu/fmti/December-2025/index.html). It would've been 100% were it not for 5 indicators, 'External reproducibility of capabilities evaluation', 'External reproducibility of mitigations evaluation', 'Geographic statistics', 'Enterprise users' and 'AUP enforcement frequency'. They disclosed how much CO2 was emitted for a model, for example: "295.2 tCO2eq for Granite 3.0 8b.""The resolution also notes that contributors are barred from feeding confidential or security-sensitive Debian material, embargoed bug info, credentials, or infrastructure details into third-party AI services without authorization, and reiterates that large-scale automated actions still require prior discussion and a human who stays accountable for the result."Obvious, nobody wants to shove confidential stuff into an AI, but there arw morons here and there that mistakenly include private API tokens or stuff that should remain secret into ChatGPT or Claude or whatever.