GNU social JP
  • FAQ
  • Login
GNU social JPは日本のGNU socialサーバーです。
Usage/ToS/admin/test/Pleroma FE

  • Public

    • Public
    • Network
    • Groups
    • Featured
    • Popular
    • People

Embed Notice

HTML Code

Corresponding Notice

  1. Embed this notice
    Charlie Stross (cstross@wandering.shop)'s status on Monday, 10-Mar-2025 06:39:28 JSTCharlie StrossCharlie Stross

    It occurs to me that Ken Thompson's classic lecture "Reflections on Trusting Trust" has a LOT to say about code generation using LLMs.

    The Thompson self-propagating UNIX hack that he described was a lot like a prompt injection attack (the right parsed input could trigger an unexpected output via opaque-to-programmer intermediate steps).

    https://www.cs.cmu.edu/~rdriley/487/papers/Thompson_1984_ReflectionsonTrustingTrust.pdf
    https://mastodon.social/@gmh/114132067525074476

    In conversationMonday, 10-Mar-2025 06:39:28 JST from wandering.shoppermalink

    Attachments

    1. No result found on File_thumbnail lookup.
      Gideon Hallett (@gmh@mastodon.social)
      from Gideon Hallett
      @cstross@wandering.shop Something that strikes me is that news disinformation is a relatively minor threat here; there's enough news around for fact-checking AI-boosted propaganda. I think a bigger threat is AI-boosted unsafe programming. Too many people rely on Copilot or GPT to write their code for them, perform no sanity-checking, and put it into workplace production. If you pollute a LLM with backdoored code, how many people will roll it out? The phrase 'Word macros for the 21st Century' springs to mind.
  • Help
  • About
  • FAQ
  • TOS
  • Privacy
  • Source
  • Version
  • Contact

GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.

Creative Commons Attribution 3.0 All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.