• 14 Posts
  • 1.01K Comments
Joined 1 year ago
cake
Cake day: June 15th, 2023

help-circle

  • the training data is not stored in the model in its original form,

    It is not stored in the model, period. Same as you do not store the shape of the letters you’re reading right now, not even the words, but their overall meaning. Remembering the meaning of what I write here, you can then produce words and letters again and you might be close but even with this short paragraph you’ll find it very hard to make an exact replica. That’s because you did not store it in its original form, not even compressed, you re-encoded it using your own understanding of language, of the world, of everything.




  • “Oh, they didn’t. Nobody leaned the technique to paint this, there may not be any feeling behind it, or any point at all, other than ‘it looks good

    “May” being the important word, here.

    I suggest that if you cannot tell the difference between “someone who knows art did this piece” and “someone just hit generate” then you have no business critiquing art.

    And I’m pretty sure that most people could learn how to prompt successfully in a matter of days or weeks.

    …that won’t give you art skills. It’s practically impossible to develop an artistic eye, much less mind, by hitting generate, the feedback isn’t sufficient, you can’t train like that. No model prompts the same, btw, frankly speaking prompting is about the worst way to condition a model when you’re out to create something specific.

    The art is not in the fucking medium. Never was. Never will be. Come at me for this and I’ll be referencing urinals on pedestals.




  • It is illegal to use in the EU for anything even remotely sensitive. Like, if you subtitle a movie with it and it messes up noone cares, your problem, if you’re doing anything that has any legal implications, from college applications over job interviews to court proceedings, they’ll nail you to the cross. For AI to be used in such domains it has to be certified and AIs certified for even a subset of these things plainly don’t exist.

    It’s like with self-driving cars: What OpenAI is producing is pretty much on the level of Tesla’s “full self driving”. It’s not even waymo who have proper autonomy tech certified to operate in a limited area in a benevolent (to venture capital) jurisdiction (some municipality or the other). Wake me when it gets actual approval from actual regulatory bodies actively trying to break it.



  • Just skimmed his position on Ukraine and it’s the ole “only diplomacy will end the war” thing, ie. he ends up regurgitating Kremlin propaganda. Nah, Russia’s economical and political collapse will end it because Russians won’t have the collective will to get rid of Putin before that.

    He somewhat relented on the Red Khmer issue in retrospect but defended how he came to the initial conclusion, which is how he managed to repeat and repeat the same mistake again. The shit he said about Bosnia as far as I’m aware he never corrected even in parts and you shouldn’t even ever *begin" to refer to starving concentration camp inmates behind barbed wire as “thin guys”.

    I understand when USians value him for writing Manufacturing Consent (and I’ll lump Canucks into that category because broader political sphere), but there’s also a fucking reason he’s persona non grata in Europe.



  • barsoap@lemm.eetoTechnology@lemmy.worldWhat are your AI use cases?
    link
    fedilink
    English
    arrow-up
    6
    ·
    edit-2
    5 days ago

    Spaced repetition, in particular Anki with FSRS. I don’t think they advertise it as “AI” or even “ML” anywhere, but let’s just say gradient descent over gigantic datasets is involved, all to predict the time when you’re about to forget something so that Anki can prompt you just before that happens. The default predictor is generic, derived from that gigantic dataset, it’s like two handful of tuning parameters, once you’ve gone through enough cards yourself it can be tuned to your mind and habits, in particular, how you use the “hard, good, easy” buttons.

    It’s the perfect sledge hammer for the application for the simple reason that we don’t actually understand how memory works so telling the computer “here’s data from millions of med students and language learners, figure out how to predict it” is our best shot. And, indeed, it’s the best-performing algorithm even before you tune it at which point it becomes eerie.


    Relatedly, as in “no LLM, no diffusion” Proxima Fusion is using machine learning to crunch through the design space of stellerators to figure out what to prototype in the real world. Actual engineers doing actual engineering.


    Then, lastly, yes, playing around with SDXL is fun. Just make sure you can actually judge the images, developing an artistic eye by hitting generate I think is close to impossible. Definitely slower than picking up a pencil, or firing up blender and actually learning how to draw or sculpt.





  • Amazon is slower than pretty much everyone else in Germany, and it’s been like that for literal decades. To get almost universal next-day delivery nation-wide a shop needs to do exactly two things: Have the parcels ready by evening, and not be located in the absolute boondonks (which would mean two-day delivery).

    Amazon, unless when ordering via premium shipping (included in prime but not worth it for that), takes days to even pack the parcel. Then they can spend a day or two sending it from one of their logistics centres to the other until handing it over to the actual parcel service.

    What they do have going for them is the mindbogglingly huge selection. Pretty much the only upside, if you need five small things from what would be five different stores each having their order minimums for free shipping amazon is the sane choice.





  • The only reasonable way to avoid backdoors is to meticulously check the submitted code.

    Which is the job of maintainers. Which now aren’t Russian, any more. To the best of my knowledge the kernel is still accepting code from Russian citizens, ultimately not having Russians in maintainer roles isn’t going to stop the FSB from infiltrating the kernel but it certainly does make it harder.

    This also isn’t in any way a judgement on the removed people, it’s just that it so happens that if you’re a Russian citizen you’re quite vulnerable to wrench attacks. You could even say that the kernel org is protecting them from being used like that.