• 1 Post
  • 101 Comments
Joined 10 months ago
cake
Cake day: January 25th, 2024

help-circle








  • I’m not a big expert on database technology, but I am aware of there being at least a few database systems (“In-Memory”) that use the RAM of the computer for transient storage, and since RAM doesn’t use files as a concept in the same way, the data stored there isn’t exactly inside a “file,” so to speak.

    That said, they are absolutely dwarfed by the majority of databases, which use some kind of file as a means to store the database, or the contents within it.

    Obviously, that’s not to say using files is bad in any way, but the possibilities for how database software could have developed, had we not used files as a core computing concept during their inception, are now closed off. We simply don’t know what databases could have looked like, because of “lock-in.”


  • ArchRecord@lemm.eetoFediverse@lemmy.worldWhy is Mastodon struggling to survive?
    link
    fedilink
    English
    arrow-up
    5
    arrow-down
    1
    ·
    edit-2
    1 month ago

    That’s what some databases are. Most databases you’ll see today still inevitably store the whole contents of the DB within a file with its own format, metadata, file extension, etc, or store the contents of the database within a file tree.

    The notion of “lock in” being used here doesn’t necessarily mean that alternatives don’t or can’t exist, but that comparatively, investment into development, and usage, of those systems, is drastically lower.

    Think of how many modern computing systems involve filesystems as a core component of their operation, from databases, to video games, to the structure of URLs, which are essentially usually just ways to access a file tree. Now think of how many systems are in use that don’t utilize files as a concept.

    The very notion of files as an idea is so locked-in, that we can rarely fathom, let alone construct a system that doesn’t utilize them as a part of its function.

    Regardless, the files example specifically wasn’t exactly meant to be a direct commentary on the state of microblogging platforms, or of all technology, but more an example for analogy purposes than anything else.

    What social media platforms don’t have some kind of character limit?

    What platforms don’t use a feed?

    What platforms don’t use a like button?

    What platforms don’t have some kind of hashtags?

    All of these things are locked-in, not necessarily technologically, but socially.

    Would more people from Reddit have switched to Lemmy if it didn’t have upvotes and downvotes? Are there any benefits or tradeoffs to including or not including the Save button on Lemmy, and other social media sites? We don’t really know, because it’s substantially less explored as a concept.

    The very notion of federated communities on Lemmy being instance-specific, instead of, say, instances all collectively downloading and redistributing any posts to a specific keyword acting as a sort of global community not specific to any one instance, is another instance of lock-in, adapted from the fediverse’s general design around instance-specific hosting and connection.

    In the world of social media, alternative platforms, such as Minus exist, that explore unique design decisions not available on other platforms, like limited total post counts, vague timestamps, and a lack of likes, but compared to all the other sites in the social media landscape, it’s a drop in the bucket.

    The broader point I was trying to make was just that the very way microblogging developed as a core part of social media’s design means that any shift away from it likely won’t actually gain traction with a mainstream audience, because of the social side of the lock-in.


  • short form content with just a few sentences per post sucks.

    I 100% agree with this sentiment.

    Jaron Lanier has a great book called You Are Not A Gadget, where he talks about the way we design and interact with systems, and he has some thoughts I think reflect this sentiment very well:

    “When [people] design an internet service that is edited by a vast anonymous crowd, they are suggesting that a random crowd of humans is an organism with a legitimate point of view.” (This is in reference to Wikis like Wikipedia)

    “Different media designs stimulate different potentials in human nature.

    He talks about how when a system becomes popular enough, it can “lock in” a design, when others build upon it as standard. Such as how the very concept of a “file” is one we created, and nearly every system now uses it. Non-file based computing is a highly unexplored design space.

    And the key part, which I think is relevant to Mastodon, the fediverse, and social media more broadly, is this quote:

    “A design that share’s Twitter’s feature of providing ambient continuous contact between people could perhaps drop Twitter’s adoration of fragments.

    Fragments, of course, meaning the limited, microblogging style of communication the platform allows for. I’ve seen some Mastodon instances that help with this, by not imposing character limits anywhere near where most instances would, opting for tens of thousands of characters long. But of course, there is still a limit. Another design feature by Twitter that is now locked in.

    But of course, people are used to that style of social media. It’s what feels normal, inevitable even. Changing it would mean having to reconceptualize social media as a concept, and might be something people aren’t interested in, since they’re too used to the original design. We can’t exactly tell.

    As Lanier puts it,

    “We don’t really know, because it is an unexplored design space.”




  • The problem is that welfare systems, such as those that provide housing, that distinguish who is eligible by how much they can afford it, to a certain degree, inevitably depress higher levels of economic activity, and good saving behavior, through the very nature by which they’re operated.

    If we say that someone is no longer eligible for free housing if they earn, say, $2,000+ a month, and housing would otherwise cost $500 a month, then if they’re currently earning $1,500 a month (the same they’d effectively have if they had to pay $500 a month for housing on a $2,000 a month salary) they have a direct incentive to not make over $2,000, unless they can guarantee they’ll make at least that much plus $500 more to compensate for the difference. If they earned $1,800 a month, they’d be making $300 more than someone making $2,000, but paying $500 a month for housing after hitting the cutoff.

    This isn’t just a hypothetical either. While this 2021 study does mention some benefits of means testing, such as more targeted expenditure, it ultimately shows that…

    “An asset means-test incentivizes low-income households to hold few financial assets making them vulnerable to predictable and unpredictable income changes.”

    …and sees that, in the end, while it can marginally increase the cost of these social programs to the taxpayer, it ultimately does more to benefit the individuals receiving the assistance.

    Or how about this research done by the Cleveland branch of the Federal Reserve that states:

    “the elimination of testing limits, such as in policies similar to a UBI, could present a welfare-improving alternative to the current system, though not without large economic trade-offs.” (They effectively mean worse targeting of funds, but better overall results)

    Means-testing directly reduces the incentives that lead to higher overall household wealth, and quality of life.

    Not having means-testing increases total income, which also means increased tax revenue. That same tax revenue can then go to funding the housing system as a whole, but it won’t directly, substantively punish people for an increase in income past an arbitrary threshold.

    Not to mention the increased administrative cost of performing means testing, as opposed to doing unconditional support, which could reduce the amount of money actually going to funding housing, in favor of funding jobs for people that audit income levels of housing applicants.


  • Anything we humans need for fundamental existence in today’s society should be free to the individual, and be a cost we all pay as a society to respect the existence of other human beings. Anything above that is up to the individual to either provide for themselves, or receive as a result of the value they contribute to society through labor.

    That’s my broader belief system, and thus, housing falls under that for me. The better we meet individual’s needs, the easier it becomes for them to contribute back to society, and experience upward class mobility.

    I believe that if we are to make housing a right, we can’t even just say it’s a right “unless you have no job,” or “unless you’re unable to fork over $500 a month,” because employment is ultimately up to the discretion of employers, who, even today, don’t even consider most unhoused people for jobs, because they don’t have stable housing, but to get stable housing, those people need jobs. (this even applies to many shelters, which will require unhoused people to either be employed, or be constantly seeking employment)

    We know that adding hoops to jump through to get welfare assistance only harms those who need to depend on it the most, without providing any significant socioeconomic benefit, so why should we apply that same logic to housing, if we determine that it should be a right of all human beings?

    I’m not saying the housing has to be great. It doesn’t need to be spacious, have all the amenities, or even have things like good quality lighting or good soundproofing from adjacent housing units, but at a bare minimum, everyone deserves somewhere to live.






  • It’s a bit more nuanced than that, because a human can still develop artistic skills by observing non-artistic creations beforehand.

    For instance, the world’s very first artist probably didn’t have any paintings or sculptures to build off.

    I’m not saying I necessarily agree that the person isn’t an artist because they rely on external training data, but generative AI models most certainly need to observe other works to ‘learn’ how to make art, whereas humans don’t necessarily have to. (Although if someone were to make a reinforcement learning model based on user feedback as a way to entirely generate better and better images starting from random variation, that would make the original training data point moot)