• 0 Posts
  • 56 Comments
Joined 2 years ago
cake
Cake day: June 10th, 2023

help-circle

  • I’d summarize the current OS situation as

    Windows Just Works until it doesn’t, at which point there’s basically nothing you can do about it and you just have to kick it until something clicks into place and it starts working properly again.

    Whereas linux Just Works to a slightly smaller degree, but when it stops Just Working it does so in granular steps most of the time, and every part of the ecosystem tries to help you fix things when they break.

    Windows is a resin-potted black box that takes input and does stuff, if it breaks you’re supposed to just chuck it and buy a new one.
    Linux is a slightly bulkier thing that you can just unscrew and replace a capacitor when it breaks.


  • part that, and part just that windows is successful because it’s successful. Everyone learns windows, thus everyone uses windows, thus everyone teaches windows.
    It’s like how all life on earth produces and consumes a specific form of sugar, but when you make sugar in a lab you get both forms, and the second form is completely inert to all digestive systems on earth.










  • sure a model could theoretically be open source, but in this case that source would be 1) the software to train the model and 2) the training data, and good fucking luck getting either of those honestly.

    the AI sphere has co-opted the term “open source” because it sounds nice, what they actually mean is just “a free copy”, a free copy of deepseek is available for download.

    They seem to be able to fiddle a lot with the models, but in the end calling it open source is just blatantly not fucking true.


  • I think i have 3 big criteria:

    • Track record
    • Structuring things to pre-emptively keep themselves (and more importantly, those who might take over later) honest and aligned with the collective good
    • Good people involved and ideally in charge of the project

    Other people have mentioned things like venture capital and that’s certainly something to bear in mind (arguably part of the structure), but there are projects like Matrix where that feels quite marginal to me, the aforementioned aspects more than make up for it.
    Like when the main figurehead of the project goes on stage and nerds out about the code, that’s a pretty fucking good sign in my book.





  • My money is on IPFS, because it’s so simple (like, in principle, obviously it’s complex under the hood). It’s not fancy, it’s basically a better version of torrent and only handles static data, but it does that really fucking well.
    It takes any data you add to your node, splits it into small blocks, does a fancy hash of those blocks, and then builds a tree of pointers that point to pointers that point to the constituent blocks. This means that any identical blocks have the same address, and thus only need to be sent once! And the same goes for anything that ends up being identical in structure, it has the same address and only needs sending once, and if for example two people rip a copy of the same obscure DVD and host it on a node, they will both provide the data to downloaders despite never having interacted with each other at all!

    This is of course massively boner-inducing for anyone who cares about archiving stuff.

    In effect it does the same thing that HTTP or FTP or whatever does, but in a modern and fundamentally decentralized way. You don’t care where the data comes from, you just request the ID from whatever nodes you can see, if they don’t have it they forward the request to those they can see etc etc, if anyone has it they reply to you and start sending the data, and then you do some fancy math to verify that it’s correct.


  • Basically everything from https://www.protocol.ai/work, bar the blockchain stuff which i can only assume they’re doing to milk investors to fund the actually good projects.

    libp2p abstracts away networking so you can simply point to a peer ID and the computer figures out how to connect to it (though you can of course specify how to connect if you wish), and it preserves connections across different networks among other stuff.

    IPFS is basically just torrent but better in every way, foremost in that you can just slap some data onto an IPFS node and if anyone else happens to be serving the same data then you’ll both automatically be valid providers, despite never having talked to each other in any way. No more needing to search out a magnet link, just seed whatever data you want and anyone requesting the data from the network will automatically find you.