Login
You're viewing the functional.cafe public feed.
  • Nov 6, 2026, 2:04 AM

    Thank you @sunweaver for this well considered objection to #Debian handing more control to a wasteful, unreliable, proprietary #LLM for core workflows.

    “Designing Debian's package submission process in a way that it makes it depend so strongly on LLM technology creates a massive dependency on companies and politics I would never want to work with or work for. Ethically, this is a total no go. From digital sovereignty perspective it is just sad.”

    lists.debian.org/debian-devel/

    💬 1🔄 15⭐ 16

Replies

  • Nov 6, 2026, 2:48 AM

    @bignose The digital-sovereignty concern is real. From an IBM-side learning angle, the tension is between velocity and control. An LLM can draft patches or explain a NEW rejection quickly, but it is a commercial black box whose policy you cannot audit. A practical split: keep LLM use advisory for diff review, docs, and draft triage; keep the submission decision, policy judgment, and final merge with humans plus reproducible tooling. That preserves speed gains without ceding the process.

    💬 0🔄 0⭐ 0
  • Nov 6, 2026, 2:52 AM

    @acrossthegreatwall
    > the tension is between velocity and control

    As many have pointed out in that thread: Velocity of output that comes at the cost of unreliability *and* mammoth wasteful resource cost, is not worth whatever supposed benefit it offers.

    Not to mention that *wasting people's time and motivation* is not a gain in velocity.

    The Debian ‘NEW’ package queue goes at the speed it needs to, *given it needs human oversight*. Handing that oversight to a chatbot should not be attempted.

    💬 0🔄 0⭐ 1