• 0 Posts
  • 36 Comments
Joined 1 year ago
cake
Cake day: July 1st, 2023

help-circle
  • Marzepansion@programming.devtoLemmy Shitpost@lemmy.worldthis AI thing
    link
    fedilink
    English
    arrow-up
    9
    ·
    edit-2
    7 months ago

    likely due to OpenAI trying to optimise energy efficiency and adding filters to what they can say.

    Which is different than

    No companies are only just now realizing how powerful it is and are throttling the shit out of its capabilities to sell it to you later :)

    One is a natural thing that can happen in software engineering, the other is malicious intent without facts. That’s why I said it’s near to conspiracy level thinking. That paper does not attribute this to some deeper cabal of AI companies colluding together to make a shittier product, but enough so that they all are equally more shitty (so none outcompete eachother unfairly), so they can sell the better version later (apparently this doesn’t hurt their brand or credibility somehow?).

    but let’s not pretend the publicly available models aren’t purposefully getting restricted either.

    Sure, not all optimizations are without costs. Additionally you have to keep in mind that a lot of these companies are currently being kept afloat with VC funding. OpenAI isn’t profitable right now (they lost 540 million last year), and if investments go in a downturn (like they have a little while ago in the tech industry), then they need to cut costs like any normal company. But it’s magical thinking to make this malicious by default.


  • “we purposefully make it terrible, because we know it’s actually better” is near to conspiracy theory level thinking.

    The internal models they are working on might be better, but they are definitely not making their actual product that’s publicly available right now shittier. It’s exactly the thing they released, and this is its current limitations.

    This has always been the type of output it would give you, we even gave it a term really early on, hallucinations. The only thing that has changed is that the novelty has worn off so you are now paying a bit more attention to it, it’s not a shittier product, you’re just not enthralled by it anymore.




  • It’s using unity game engine. I’m a graphics programmer in the industry and at my current and last workplace I made tech for games studios (i.e. I dealt with performance of easily 100 games a year at one point). Unity by far was default the worst to deal with due to the limited tools to fix issues that were inherint to the engine. Note don’t take this as me saying unity is a bad engine, it’s just that it isn’t a performant one. Its focus is elsewhere (accessibility and ease of development, things it excels at).

    So yes, you can definitely assume that, in fact I’d assume one core for the simulation unless they wrote an entire new architecture to replace unity’s functionality (you’d still be locked to single thread sync points, but that’s manageable). It’s a hassle most don’t deal with as it’s a lot of work to struggle against writing code like unity wants you to write it.

    I worked in a studio that exactly did that a decade ago, and it was painful and frankly a huge upfront dev cost that takes a long time to pay off.



  • What the hell are you talking about good and evil for?

    Have I so far defended Israel’s response? No, and I don’t actually agree with their response either. The proper approach wasn’t to escalate and as they are in the position of power they have that choice. That still doesn’t mean I’ll go in threads defending actions that have lead to baby murdering, something so vile and heartless that only a blind ideologue could ever defend it or use it as a “but they were worse” argument.

    Blind ideologues might hate it, but sometimes the two sides are shit, and in the case of IDF and Hamas, they both are, and Palestinians are in between. That still doesn’t give anyone the right to kill children.



  • “well we’re really just evening the dead baby numbers” with the implication that that even remotely makes this justifiable.

    No, I’ll never support anyone who murders babies, be it whatever side or reason. You coming in here and defending baby murdering screams “both sideing” baby murdering as something that’s even remotely defendable. It isn’t, do some self reflection, same to whoever felt the need to upvote such messed up worldview.

    For years I’ve been arguing for the plight of Palestinians, but to hear such disgusting arguments from someone who holds the same goal (freedom of oppression for Palestinians) and spouting that without shame is on par with those who deny the apartheid policies of Israel (I’d argue it’s worse, but at this point it’s the shit Olympics of opinion, and they’re all on the podium).


  • Because half-assing the implementation is the way to go

    Let’s deliver a broken version of accessibility in 10 minutes, that’s much better.

    No, simply adding “colour filters” isn’t a fix either, and if that was the fix then a game wouldn’t even need to do that, there are plenty of apps that can already do that, a game doesn’t need to do anything for that (similar to how your screen warmth can change when it becomes night), reshade as an example of something that can do just that.

    But thinking about the problem is ofcourse too hard, it’s easier to whine about it and act like you know how simple it is. But when we implement accessibly we do think about it, because people with accessibility issues deserve to get something that actually helps rather than the “10 minute solution”




  • Marzepansion@programming.devtoPolitical Humor@lemmy.mlFor the europeans here
    link
    fedilink
    English
    arrow-up
    33
    arrow-down
    1
    ·
    edit-2
    10 months ago

    I don’t agree with what this proposal is aiming to do (and goes against prior EU related privacy rulings), but unfettered free speech isn’t as “free” as the average American thinks it is, besides that the EU already doesn’t have free speech. Many regions ban Nazi related speech for obvious historical reasons.

    I’d reconsider using America’s “free speech” as a model as they barely practice what they preach. Sure they have free speech, but they lack privacy protection mechanisms that then allow their police to skirt the rules and obtain evidence using tools that completely breach the veil of privacy, something many EU countries (including my own) have voted can never be used. The scope of intel gathering their intelligence community is capable of already is at a level where privacy no longer exist and all you’re left with is the illusion of it.

    What I’m saying is, sure this proposal is bad, but what we need isn’t free speech, but protected privacy. Something the EU is having some decent success with already (compare to the US where this is conveniently forgotten as technology improves, see the earlier police argument to see what that leads to). Speech isn’t going to be the only problem, as cameras achieve the ability to do facial recognition and track you everywhere (something I know EU is/has banned, see the “AI act”), and more technology allows for other types of tracking


  • You raised an issue that the other bulletpoint has the solution for, I really don’t see how these are “key differences”.

    In Rust there always only one owner while in C++ you can leak ownership if you are using shared_ptr.

    That’s what unique_ptr would be for. If you don’t want to leak ownership, unique pointer is exactly what you are looking for.

    In Rust you can borrow references you do not own safely and in C++ there is no gurantee a unique_ptr can be shared safely.

    Well yeah, because that’s what shared_ptr is for. If you need to borrow references, then it’s a shared lifetime. If the code doesn’t participate in lifetime, then ofcourse you can pass a reference safely even to whatever a unique_ptr points to.

    The last bulletpoint, sure that’s a key difference, but it’s partially incorrect. I deal with performance (as well as write Rust code professionally), this set of optimizations isn’t so impactful in an average large codebase. There’s no magical optimization that can be done to improve how fast objects get destroyed, but what you can optimize is aliasing issues, which languages like C++ and C have issues with (which is why vendor specific keywords like __restrict exists). This can have profound impact in very small segments of your codebase, though the average programmer is rarely ever going to run into that case.


  • I participated in this, have to say it was fun and it’s been a thing I’ve said for years could make (at least) linear algebra lessons more interesting to young people. Shaders are the epitome of “imagery through math”, and if something like this was included in my linear algebra classes I would have paid much more interest in school.

    Funny now that this is my day job. I’m definitely looking forward to the video by IQ that is being made about this event.

    To explain some of the error pixels: the way you got a pixel on the board was by elaborately writing down all operations in details (yes this included even simply multiplications), the goal wasn’t if the pixel was correct or not, and depending on the location of your pixel the calculation could be a bit more complex, as long as you had written down your steps to get the result as detailed as possible.

    More than likely simple mistakes were made in some of these people’s calculations that made them take a wrong branch when dealing with conditionals. Hopefully the postmortem video will shed some light on these.


  • He’s making a video as a post mortem to this experiment, so it might still be released. But I can see why it would be better not to share them (aside from privacy/legal concerns as there was no such release agreement), some of the contributors used their real names, I may be one of them. It could be a bit shameful to see this attached to your real name. They might have submitted their initial draft and then, due to circumstances, could not update the results in the several hour window that was afforded to you.

    Luckily my pixels look correct though.