Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Debian votes to let contributors code with AI

2»

Comments

  • LEBUserJoeLEBUserJoe Member
    edited September 1

    @forest said:

    @LEBUserJoe said: suspect this is mainly due to the limitations of truly good hunting models being limited from public use currently.

    But now that Kimi K3 is out...

    Even Kimi K3 although truly destructive if you have access to enough usage, it's still nothing to internal OpenAI, and Claude models used purely for penetration testing.

    I hate there benchmark but it's someything

    "ExploitBench, K3 scored 32%, above GLM-5.2’s 24%."

    That that got given access to mythos for example achieved "78%" on the exact same benchmark, and this is still technically a "public model" they have even better in-house same with OpenAI.

    My point was more, they are being used for the top software, top projects. It will be devastating to a lot of projects / older running software once they are targeted too even by a Kimi K3 model, but they are truly dog for reversing / penetration testing comparatively.

    Wait until these level of models, are in the hands of people who are not willing to KYC and prove themselves to gain access, once this level is open-source the attack surface is literally billions X fold.

  • jsgjsg Member, Resident Benchmarker

    @forest said:

    @jsg said:

    @forest said:

    @jsg said: Btw, how is quantum computing doing? *g

    Improving at an unsettling pace to the point that I'm trying to cover my ears and convince myself that Gutmann has a point.

    *g

    I don't know what *g means.

    It means "grin(ning)"

  • stable_geniusstable_genius Member
    edited September 1

    @forest said:

    @jsg said: Btw, how is quantum computing doing? *g

    Improving at an unsettling pace to the point that I'm trying to cover my ears and convince myself that Gutmann has a point.

    As Sundar Pichai said in 2025:

    practical quantum computers are at least five to ten years away

    I can confirm that, we've been hearing that statement for about 20 years now so it must be correct. I'm fully confident that it will stay correct for many decades to come.

    Thanked by 1jsg
  • As other have said, it would be very difficult to enforce a "no-AI rule" at a project level, as how does one prove they didn't use AI. (Submit git commit logs, show proof of progress, etc.) Where does one draw the line? Is AI code completion okay?

    I have never worked on a project of Debian's size, the reality is that if they outright banned AI use they would then have to enforce it which means either having some code litmus tests to try to determine which depending on the AI, size of the commit, etc. could be easy or hard. But at the end of the day I want the open source devs focused on creating new features, solving the problem, and working to deliver quality code. If the code reviewer looks at the code and doesn't _waste _ time trying to determine if it is was AI or not, and the quality gates still exist for releases, especially stable, etc. I don't have a problem with it.

    If we suddenly start seeing 3,000+ lines of code changes that don't fit the quality/coding practices, breaking changes, etc. then that is a different story and I would check out other distros.

    Thanked by 1jkl9
  • Working code cannot be hallucinated because it works - same as delusion stops being a delusion the moment it becomes true.

    Paranoia over AI code is absurd when its on you to write tests and we have major more important social problems with AI.

  • forestforest Member
    edited September 1

    @stable_genius said: I can confirm that, we've been hearing that statement for about 20 years now so it must be correct. I'm fully confident that it will stay correct for many decades to come.

    There's no way to know when it will happen, but there have been a constant stream of steady breakthroughs.

    The measure the layperson is using, such as "size of integer factorized with Shor's algorithm", is not a very good one. Sure, the best it's done is factor a two digit integer so far, but once it can factor (say) a 128-bit integer (a task that can be done trivially with the classical algorithm GNFS on any old laptop), scaling that up to 4096 bits or more is comparatively easy.

    And it's awfully close to being able to do that with all the advances in quantum error correction.

    Regardless, we should be switching to hybrid PQC.

  • The end is nigh?

  • stable_geniusstable_genius Member
    edited September 2

    @Andy10tbit said:
    The end is nigh?

    The end of the beginning that's what it is.

    It is also a beginning, the beginning of the end.

Sign In or Register to comment.