• 2 Posts
  • 731 Comments
Joined 2 years ago
cake
Cake day: June 6th, 2023

help-circle
  • I agree that you can’t know if the AI has been deliberately trained to act nefarious given the right circumstances. But I maintain that it’s (currently) impossible to know if any AI had been inadvertently trained to do the same. So the security implications are no different. If you’ve given an AI the ability to exfiltrating data without any oversight, you’ve already messed up, no matter whether you’re using a single AI you trained yourself, a black box full of experts, or deepseek directly.

    But all this is about whether merely sharing weights is “open source”, and you’ve convinced me that it’s not. There needs to be a classification, similar to “source available”; this would be like “weights available”.



  • Is there any good LLM that fits this definition of open source, then? I thought the “training data” for good AI was always just: the entire internet, and they were all ethically dubious that way.

    What is the concern with only having weights? It’s not abritrary code exectution, so there’s no security risk or lack of computing control that are the usual goals of open source in the first place.

    To me the weights are less of a “blob” and more like an approximate solution to an NP-hard problem. Training is traversing the search space, and sharing a model is just saying “hey, this point looks useful, others should check it out”. But maybe that is a blob, since I don’t know how they got there.



  • Yeah, I agree that in the long term those two sentiments are inconsistent, but in the short term we have to deal with allegedly misguided layoffs, and worse user experiences, which I think makes both fair to criticise. Maybe firing everyone and using slop AI will make your company go bankrupt in a few years, and that’s great; in the meantime, employees everywhere can rightfully complain about the slop and the jobs.

    But yeah, I don’t think it’s fair to complain about how “inefficient” an early technology is and also call it “magic beans”.




  • Yeah, I understand that you personally choose to disagree with reality, maybe you don’t like what reality has become, but unfortunately that doesn’t make it less real.

    Twitter wasn’t profitable for its entire existence, it’s often a cesspool of ragebaiters, but clearly it has value because the second it was taken over, everyone insisted on continuing to use it, even choosing to migrate to various clones.

    Uber and Lyft have been struggling to be profitable by effectively stealing from their drivers, but millions of people get off a plane and immediately use the services every day. It clearly has value.

    Same for doordash and uber eats.

    Your personal distaste for the business practices are valid, but they’re not relevant when discussing what the current state of the technology is. For many millions of people, chatgpt has (for better and worse) replaced traditional search engines. Something like 80% of students now regularly use AI for their homework. When Deepseek released, it immediately jumped to #1 on the Apple Store.

    None of that is because they’re “magic beans” from which no value sprouts. Like it or not, people use AI all. the. time. for everything they can imagine. It objectively, undeniably has value. You can staunchly say pretend it doesn’t, but only if you are willingly blind to the voluntary usage patterns of hundreds of millions (possibly billions) of people every hour of every day.

    And for the record, I am not in that group. I do not use any LLMs for anything currently, and if anything makes me use AI against my will, I will promptly uninstall it (pun intended).




  • Lol this article is very relevant to a lot of scam industries (essential oils, Earthing, 5G protection crystals, etc), but AI is objectively not one of them.

    Regardless of how much of a bubble we’re in, regardless of how many bad ideas are being pushed to get VC funding or pump a stock, regardless of how unethical or distopian the tech is, AI objectively has value. It’s proving to be the most disruptive tech since the world wide web (which famously had a very similar bubble of bad ideas), so to call it “magic beans” is just wishful thinking at best.










  • Yeah, so the best beginner resource (especially for old hardware) is honestly this old blog series A Trip Through the Graphics Pipeline. But importantly, it was written before modern dx12/vulkan were around, so it will use dx9 terminogy. Also, it’s possible that certain aspects of the vulkan api aren’t possible to faithfully implement to spec with older hardware (while still maintaining reasonable performance, or possibly at all).

    From there, it’s probably best to try and implement a backend for your GPU into the radv mesa driver, so probably go take a look at how other GPUs are done.

    You will need to become familiar with the AMD GPU programming docs, here.

    I’m not going to be one of the naysayers here who says you shouldn’t even try to do this, but as an ex-graphics driver dev, I think you will find pretty quickly that you have your work cut out for you. It would probably be easier to implement a Vulkan-on-OGL translation layer.

    Both GPU hardware and drivers are developed by teams of professionals, each of which is hyper specialized in a few components, because none have the time or ability to be familiar with everything about all components (at least, not while also being effective). I’m not saying you can’t do it, and I’m not saying you wouldn’t learn a LOT doing it, but I am saying that by the time you finish, you could have worked a minimum wage job and purchased a dozen 5090s 😉.

    Edit: and oh right, then there’s reading through the Vulkan spec, which, if it’s your first graphics API, will take months, if not years to digest.