• 0 Posts
  • 724 Comments
Joined 2 years ago
cake
Cake day: June 11th, 2023

help-circle


  • Unless we invent cold fusion between the next 5 years, they will never be economical. They are the most energy inefficient thing ever invented by humanity and all prediction models state that it will cost more energy, not less, to keep making them better. They will never be energy efficient nor economical in their current state, and most companies are out of ideas on how to shake it up. Even the people who created generative models agree that they have just been brute forcing by making the models larger with more energy consumption. When you try to make them smaller or more energy efficient, they fall off the performance cliff and only produce garbage. I’m sure there are researchers doing cool stuff, but it is neither economical nor efficient.


  • dustyData@lemmy.worldtoTechnology@lemmy.worldruh roh
    link
    fedilink
    English
    arrow-up
    5
    ·
    8 days ago

    Several tried. Nothing as elaborate as cross dissemination, federation or whatever. But at least 5 to 10 years ago it proved to be almost impossible. Platforms like Rooster teeth, which was 100% subscription based, I think never broke even and still relied on YT ads for the majority of the revenue. Some big and small channels tried to at least just catalog, archive and serve their own videos and the costs still became astronomical really fast. Whenever you see one of those very old channels, most of them don’t conserve copies, let alone original source footage of their entire material. Everyone just delete their videos once they’ve been on YouTube for a month or so now, and they have to download their own videos when they want to reuse old footage.

    Storage is cheap today, yes, but video really eats storage at an alarming rate. Specially now that 4k is the standard. So you have to reuse storage over and over. Transcoding is also really fast and optimized with modern algorithms, but it takes specialized graphical cards and data centers charge a premium to use servers with such capacities. Self hosting will never be able to satisfy a moderate demand. Get anything above 100 users simultaneously transcoding videos and a non-specialized server will halt to a grind just on IO calls to hard drives alone.

    Once you consider all those factors it is obvious why YouTube is such a miracle.






  • Indeed, the formal definition actually doesn’t specify how many monkeys will write what given an infinite number of monkeys, it’s unknowable (that’s just how probability is). We just know that it will almost surely happen, but that doesn’t mean it will happen an infinite amount of occurrences.

    The infinite amount of time version is just as vague, one monkey will almost surely type a specific thing, eventually, given infinite time to type it. This is because when you throw infinites at probability, all probabilities tend to 1. Given an infinite amount of time, all things that can happen, will almost surely happen, eventually.




  • dustyData@lemmy.worldtoTechnology@lemmy.worldThe GPT Era Is Already Ending
    link
    fedilink
    English
    arrow-up
    20
    arrow-down
    1
    ·
    20 days ago

    The whole point is that one of the terms has to be infinite. But it also works with infinite number of monkeys, one will almost surely start typing Hamlet right away.

    The interesting part is that has already happened, since an ape already typed Hamlet, we call him Shakespeare. But at the same time, monkeys aren’t random letter generators, they are very intentional and conscious beings and not truly random at all.



  • Continuing this take. From a storytelling point of view, they should’ve made it so that having a lightsaber was extremely difficult, the defining feat of a master Jedi knight. Something that padawans trained to use eventually but was an actually really hard, life threatening even, object to create. Crystals should’ve been an statistical impossibility, involve a pilgrimage and ceremony, you’d have to be a keen user of the force, train your sensibility to it, master the skill of manipulating life and matter through the force to construct it. Sabers had to be relics, with names, history and mythology. Handed from master to padawan when they became knights through the ages. Further symbolizing the master-apprentice relationship. Thus there can’t be any more apprentices than there are masters. Sith would have to kill Jedis and steal them, corrupting the sabers.

    But Lucas was a meh world builder anyways, so whatever.




  • The only money to be made in the LLM craze is data scraping, collection, filtering, collation and data set selling. When in a gold rush, don’t dig, sell shovels. And AI needs a shit ton of shovels.

    The only people making money are Nvidia, the third party data center operators and data brokers. Everyone else running and using the models are losing money. Even OpenAI, the biggest AI vendor, is running at a loss. Eventually the bubble will burst and data brokers will still have something to sell. In the mean time, the fastest way to increase model performance is by increasing the size, that means more data is needed to train them.



  • Oh there are these things I call Hollywood wounds. All these wounds that are supposed to be survivable in action movie land but would be very real life threats. Stabbings are nasty, specially in the abdomen. You just got signed to play craps with loaded dice and the casino is abdominal cavity infections. You might survive the bleeding, but the fever can still take you out several days later.

    There are others like being shot in the leg or the shoulder. Technically survivable, but will destroy your mobility forever. Another one is vertebral spine fractures (which this batman also miraculously survives). If you can walk again after that it would be after years of rehab, and even then you’d still be in pain the whole time until the day you die. Superhero movies are very unrealistic on their handling of human anatomy.