cultural reviewer and dabbler in stylistic premonitions

  • 236 Posts
  • 540 Comments
Joined 5 years ago
cake
Cake day: January 17th, 2022

help-circle





  • Arthur Besse@lemmy.mltoOpen Source@lemmy.ml•Ban AI Slop?
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    5 days ago

    I don’t really want to become the judge of what is “heavily-AI-assisted”.

    imo it doesn’t really take much in the way of judgement; usually it is pretty obvious when there is one developer and the README is written in LLM voice and the project is a few weeks or days old.

    I think if I were to write a rule I would focus more on self-promotion in general than trying to decide how AI assisted it was

    I think that would also be a good rule, but wouldn’t stop people from promoting others’ slop or from promoting their own while pretending it isn’t.

    I would generally rather rely on user votes to decide if it is “worth it” or not

    unfortunately there are sometimes cases (such as the post i deleted here yesterday) where a post will have mostly upvotes despite being not-even-open-source. there are also things which are freely licensed, but get many upvotes despite being absurdly bad vibe-coded ideas - eg, this centralized IPFS web directory making contradictory privacy claims 🫠 (imo that should be deleted; what do you think?)


  • Arthur Besse@lemmy.mltoOpen Source@lemmy.ml•Ban AI Slop?
    link
    fedilink
    English
    arrow-up
    4
    ·
    edit-2
    5 days ago

    Curious. Woke up to find my post removed at almost exactly the same time you posted this, complete with an accusation of “slop”.

    yes, after a brief assessment i concluded (correctly, i see now after taking another look at it) that your project is exactly the kind of new, single-developer, vibe-coded slop which i think we should not accept the promotion of here… and yes, seeing and deleting your post was what reminded me that i’ve been meaning to post something in this thread.

    however, after i deleted it for being slop, i also noticed another reason it is not suitable for this community: as @thingsiplay@lemmy.ml helpfully pointed out in the comments, your README says this:

    To that end, I’m publishing the current 32.71 source code and final release, with the source code frozen at this version.

    Download the APKs from the SmartTube LOCAL GitHub Releases page.

    I’ll continue to publish updated APKs as I hatch them.

    So, it isn’t even an open source project going forward: the released source code is “frozen” and future versions will be binary-only? so, in this case you actually shouldn’t post this in the !aials@lemmy.ml community either 🤦

    fwiw i am personally not completely opposed to AI-assisted coding; i am also doing some of it myself. however i think everyone should be skeptical of new single-developer vibe-coded projects - most are very low quality and will be abandoned in the near future.

    The latter leads to things like this.

    This is exactly what the new community is meant to address: to have a place for people to share their AI-created softwares without getting a flood of hate for it being made with AI.


  • Arthur Besse@lemmy.mltoOpen Source@lemmy.ml•Ban AI Slop?
    link
    fedilink
    English
    arrow-up
    2
    ·
    6 days ago

    imo this community should have a rule against posts promoting new primarily-written-with-an-LLM software projects. as an admin, i’ve deleted a few myself when there are reports of them and the mods aren’t resolving them.

    banning all AI-assisted software is obviously no longer reasonable now that many large and old projects are using it (with varying levels of care…) but I don’t think we actually need a very firm definition of “slop” to be able to say some things clearly are slop and some AI-involved things clearly aren’t. When in doubt about if something is of sufficient quality that we should keep it here, the age of the project and the number of developers involved are useful secondary metrics.

    @gary_host_laptop@lemmy.ml created a new community which I think should be linked in the sidebar here as an alternate place for sharing these things. How about this for a rule (in the !opensource@lemmy.ml sidebar)?

    This is not a place to promote new single-developer heavily-AI-assisted projects. Another community exists for sharing AI-Assisted Libre Software: !aials@lemmy.ml

    note that the word “slop” is intentionally absent from my proposed rule (and from name and description of the community) - the point is that this is a place for people to share their AI-created software, regardless of if it is 100% vibe coded or is actually high-quality “agentic engineering” - only time, review, and users will ultimately tell which category many of these new things fall into. But enough of it is slop that we shouldn’t have this community flooded with it, imo.

    cc @CrypticCoffee@lemmy.ml @Lettuceeatlettuce@lemmy.ml @kevincox@lemmy.ml




  • The new edition, known as TUXEDO OS “Continuous Debian,” replaces the distribution’s traditional Ubuntu foundation with Debian Testing. The change denotes a significant shift for TUXEDO OS, which is developed by German Linux hardware manufacturer TUXEDO Computers.

    […]

    Unlike Debian itself, which periodically promotes its Testing branch to Stable, TUXEDO OS will remain permanently based on Debian Testing.

    🤔

    https://www.debian.org/security/faq#testing currently says:

    Q: How is security handled for testing?

    A: Security for testing benefits from the security efforts of the entire project for unstable. However, there is a minimum two-day migration delay, and sometimes security fixes can be held up by transitions. The Security Team helps to move along those transitions holding back important security uploads, but this is not always possible and delays may occur. Especially in the months after a new stable release, when many new versions are uploaded to unstable, security fixes for testing may lag behind. If you want to have a secure (and stable) server you are strongly encouraged to stay with stable.

    I wonder if TUXEDO Computers plans to be shipping security updates themselves faster than Debian testing does?


  • LPEs are not RCEs

    They didn’t say they were, the post is very clear about the the 948 vulns they’re talking about being LPE.

    If it’s a bunch of local jvm consoles or vi escapes into shell

    if it was that, they wouldn’t be “Linux LPEs”. These are clearly kernel bugs they’re talking about.

    And again, stop taking OpenAI at their word about security.

    OpenAI is full of shit about a lot of things but I doubt they’re lying about having working exploits for these Linux bugs.




  • Something else AI’s ruining…

    can you explain what you mean by this? are they ruining linux by revealing its bugs at a much faster rate? or ruining the security research field by depriving researchers of the challenge and enjoyment of finding these bugs themselves?

    repeating myself from a previous thread: of all the things i hate about LLMs, getting vulnerabilities in important software like linux discovered and fixed is not one of them

    (although, as i said elsewhere in this thread, i find it unlikely that openai is actually disclosing all of the bugs they find 😬)





  • It is a real shame that people think that this is vibecoded

    we can’t see your source code but the copy on your transparency page sounds like LLM voice:

    • “Your feed is people and communities, not advertising.”
    • “This is a promise we keep in good faith, not a technical guarantee.”
    • “Analytics let people see how their posts, profiles and communities are doing. This is everything they record, and nothing else.”
    • “Eight promises:” […]

    If you say you wrote some of the code by hand, cool, i don’t doubt it. But it’s still giving slop.

    Also, charging money for a “plus” account on a centralized closed-source activitypub implementation is…

    blinking guy gif

    Lastly, it’s difficult to reconcile these two statements:

    as a software engineer with over 15 years of experience

    […]

    the core is closed for anti-abuse reasons

    gif meme with women talking and gesturing, subtitled "that's not how this works! that's not how any of this works!"



  • In the spirit of open-source OP asked if people want to contribute their time. You wouldn’t flame any other foss project this hard for simply being unfinished.

    asking strangers to review your LLM outputs without informing them that some (or in this case, most) of it has never been reviewed by any human before, is not at all in the “spirit of open-source”. This website continues to be very misleading: it says “Learn math and science by doing. Bite-sized, hands-on lessons. Drag, predict, and solve your way to real understanding. Free for everyone.” and has a “Start learning” button. The disclaimer which OP added on the About page absolutely needs to be moved to the front page, where everybody will see it before they see a pitch implying that this is ready to go.

    @kiarash@lemmy.world when i undeleted your post before, after you added the “less than a third” disclosure on the About page, I told you in a PM that that was insufficient and I see now you have still not fixed it so I am sorry but I am going to delete this post again. Send me a PM if/when you make the disclosure of what this actually is far more prominent. I suggest showing your agent my comments and asking it if this is reasonable and if your disclosures are sufficiently prominent.



  • Hi, i’m an admin on the lemmy.ml instance where you also posted this.

    You asked in your other post in !privacy@lemmy.ml (which I was going to delete for being privacy snake-oil, but someone else beat me to it):

    what would you want from a tool like this to trust it?

    For one thing, I would want the ability to download the whole thing (models included) in a single step, to be able to actually be sure I’m using this offline (without taking my computer offline).

    Using this kind of “nothing is uploaded, we promise” application on a website leaves the user relying on the website to keep its promise: even if the javascript has been reviewed and doesn’t usually upload a file, a user can’t ever be sure (without going offline after loading the page, which is very inconvenient and also a fragile guarantee at best given webworkers and various other browser technology) that the server isn’t sending some different javascript this time which will actually exfiltrate their secrets.

    I also don’t want an AppImage, because then I would need to trust you much more - an AppImage can do far more damage than a malicious webpage! And also I see your AppImage is only 170MB, so it obviously doesn’t include all of the models… so it itself also needs internet access.

    If what the site says is true, I should be able to host the whole thing with a local static webserver and access it using a browser which I deny internet access to (eg configure it to use a nonexistent proxy and let connections to 127.0.0.1 bypass the proxy, or, better, run it in a different network namespace where it cannot reach the internet at all). If you have some way to predownload all of the models for people to use this in an offline browser like this, you should make that clearer and easier to find.

    But, beyond that, I think you’ve also got a long way to go before you can credibly suggest this as a security and privacy tool. I’ve only had a quick glance at what you’ve built so far, and the breadth of features is certainly amazing.

    However your first (and so far only) tool which I actually looked somewhat closely at - “Secret Note Link”, source code here - is misleading users by offering them an “expiry” feature which is trivially circumventable. This is obvious to any technically savvy user who thinks for a moment about the impossibility of enforcing an expiry policy in this context (where the data is stored in the URL and not on a server), but i think the overwhelming majority people seeing this feature will not infer its pointlessness and will assume incorrectly that it actually works. This incorrect assumption is reinforced by the fact that this privacy-cosplay expiration feature is for some inexplicable reason enabled by default, with a one day expiration.

    It’s also weird to be hyping the AES aspect of your “secret link” when the default setting is to use no passphrase (which, looking at the source code, I see means that you still encrypt but with a random key which you include in the url fragment alongside the ciphertext). Without a passphrase, maintaining the secrecy of the link is all that can keep the payload secret. Encryption provides no benefit whatsoever in that case; you may as well just gzip and base64 the plaintext and put that in the fragment instead instead of using AES… or, just send the plaintext instead of sending a “secret” link at all (although given the way mobile phones work having a way to send base64-encoded text is maybe actually useful 🤡). The pointlessness of the encryption here is obvious to some people, but not to most of your target audience.

    screenshot of 'Secret Note Link' as it looked at the time this comment was written

    screenshot of website, including text "Turn a password or private message into an encrypted link that only opens in the browser, with an optional password and expiry." and "AES-256-GCM in your browser. The whole note lives inside the link, nothing is stored anywhere. Anyone who has the full link (and password) can read it, so share it only with the right person. Expiry is checked when the link is opened." and "expires after" control set to 1 day.

    I think you should remove the expiry and no-passphrase key-concatenated-with-ciphertext features altogether, and ask your agent to review all of the other tools in light of this feedback to look for and remediate similar cases of snake oil.

    I was going to remove your post from !privacy@lemmy.ml but leave the one in !opensource@lemmy.ml, but someone else removed both while I was writing this. I was going to say about the latter: I removed your post in !privacy@lemmy.ml but I’m leaving this one for now on the assumption that many of the tools that you have created are probably actually useful. fyi you should know that you’re probably going to receive a lot of negative feedback posting this here (see the discussion in the post currently pinned at the top of this (/c/opensource) community).

    I think you probably have good intentions so I am unbanning you from lemmy.ml (but not undeleting your posts to communities on this instance), and I’m posting this comment here in another community on another instance instead.

    HTH.


  • I agree. I’ll make the modifications later. Thank you.

    Thanks, glad to hear it.

    How much later, though? If it’s not soon, I think I will delete this post for now (i’m an admin on the home instance of this community). You can still edit it while the post is deleted but only you and mods and admins will be able to see it; send me a message after your edits are made and I’ll undelete it.

    To be clear what I am asking for is an early/prominent/clear disclosure, both on the website and in the body of this lemmy post, of the percentage of the lessons which you estimate have not yet been carefully read by you (or presumably anyone else).

    Thank you, and I hope this isn’t too discouraging!