Відео

If that were the case then music wouldn't exist. If everyone was always only ever copying someone else, where did it come from in the first place? Humans are capable of being derivative, but AI is only capable of being derivative. Humans are capable of being creative. AI is not. Neural networks are loosely based on what we currently think we know about how groups of neurons interact with one another. It's not the other way around. Our brains don't work like neural networks. We don't have the same limitations.

this is all mostly unfounded, you know. current AI isn't great, but there's not really any reason a priori that anything you're saying is true.

furthermore, humans aren't actually generating completely new things a priori. when one makes music it is generally informed by the music you've heard before and in your surrounding culture. sound familiar?

I disagree. As far as I am aware the notion of consonance and dissonance in its most basic form not dependant on the cultural background. For some bizzare reason our brains like when the relationship between two frequencies is a simple fraction. Sometimes the opposite is true as well. While there definitely are genres and compositions that are more of an aquired taste than others, ultimately noone taught you to like music, to enjoy it. The capacity to derive pleasure from listening to music is just something humans happen to possess inherently. The machine can not derive pleasure from listening, let alone creating. It can not wonder why a dissonant chord plucks its heartstrings in a certain deeply melancholic way. There is no self-discovery that a LLM can embark on, no new truths about the human experience it can share though a melody. There is nothing inside but weights and biases. Would you be willing to reduce yourself to weights and biases? Admit that all your life you have done nothing but be strung along by some data you happened to be trained on? If not, I hope you can understand why art and what we now call AI do not mix well.

@@koseq7 Main pro I can think of is that a pseudovector representation has less redundancy than a quaternion: A quaternion has four values; a pseudovector has three. Another positive is that they play nicely with the formula for applying torque. Torque is also often represented as a pseudovector. As for gimbal lock, that should be avoidable. Rodrigues' rotation formula lets you rotate a vector, given an axis of rotation and an angle to rotate by. All of this said though, you still probably need to use a quaternion or a matrix to represent the actual orientation of the object. IIRC matrices-based transformations tend to run slightly faster (though since this is Desmos it'll be hard to tell lol).

I'm going down a blue pill bug rabbit hole today. What's your excuse?

@@Flint-Dibble-the-Don same.