• 0 Posts
  • 22 Comments
Joined 1 year ago
cake
Cake day: August 8th, 2023

help-circle







  • Both. Everyone is afraid of AI taking over but it’s just a tool. Human augmentation is way more likely to lead there. But in the mean time, Stephen Hawking lived quite a while only being able to speak with augmentations. Just like any other technology, it will be at the very least researched in fear that someone else will first. So might as well embrace it





  • Closer, and I hope I’m not just being a pedantic jerk, but there is no code being generated either. To use correct terminology, the weights of the nodes are what change. Nodes are roughly thought of like neurons in a brain, and weights are roughly thought of as the strength of the connection between one neuron (node) and another. Real brains are way more complex.

    The weights of the nodes do contain information, but it’s not human readable at all, we actually don’t have a way of understanding how they work, just a rough idea of why. Sort of like how your brain contains the information on how to catch a ball, it performs the equivalent of calculus to do so, but there is no calculator in your brain doing the math to catch the ball. Actually, maybe a better analogy, if you have a bouncy ball, it contains the required information to bounce if you drop it, but we can’t read that information, we can only model it.

    But I’m just rambling at this point, your point is clear and valid lol


  • Sort of, but there’s no database at all, just a bunch of numbers and math. It’s almost like controlled evolution, breeding plants to select desirable traits. Except that’s another field of computing called genetic algorithms. Neural networks are a pile of math trained on data. You give it a cat, it says whether or not it thinks it’s a cat, and you tell it if it’s right or wrong, then it adjusts it’s math accordingly. Do this with a million cats and not cats and it becomes better than humans at identifying cats. LLMs are just that but with word predictions and trillions of words for training. It’s impressive in its own right


  • I don’t disagree, but I do want to point out your understanding of how chatgpt works is flawed. There is no database or query going on. It’s a giant neural network model that was trained on all that data you mentioned. The model is effectively predicting what the next word should be based on the previous words, nothing else. Each individual word is selected this way.

    It doesn’t change any of your arguments or conclusions, but I wanted to point it out, because if someone wrote a chat not like chatgpt using databases and programming I would be floored and incredibly impressed






  • I will be super annoyed. But I also just won’t buy another Kindle if they don’t have buttons. I also have a kobo with buttons that I like, so I’d just get another kobo with buttons. Though e readers don’t seem to be affected by planned obsolescence like other electronics so I won’t be buying a new one for a long time anyway.

    Wooo Amazon listen to my opinions even though I’m not going to buy your stuff!

    Sigh, I miss phones with physical keyboards.