I can definitely relate to the notion that really good ideas come from somewhere outside oneself. This happens to me sometimes with musical ideas. If I consciously try to come up with a melody, it will never sound any good - whereas any good musical ideas I've had, just sort of appear fully formed - from somewhere...
Bulgaria has a rich culture of distilling Rakia from a variety of fruits (Rakia is a type of fruit brandy). Personally I dislike the Slivi (plum) based ones. The grape is my favourite.
I do know what you mean. Everything's designed to not get damaged and not cause anyone to get sued these days. Victorian buildings and vehicles on the other hand, just seem to look gorgeous by default.
One of my favourite artists. A superb synthesis of uncanny, mystical surrealism. His work reminds me of Alfred Kubin (another favourite) - it's less grotesque, but shares the same unsettling vibe. Wonderful!
"The AI will absorb your words, and some small part of you will gain immortality. In some small but very real way, you'll live forever, some part of you ensconced safely for all eternity in a handful of vectors deep inside a pile of inscrutable matrices."
That sounds like a sort-of-religion of the future, actually.
It must cost some non trivial amount of money to run those queries through the neural networks. Charging a fee doesn't seem that unreasonable. Better would be to open the model so people could run it locally, mind you (though perhaps normal machines aren't capable of running such a massive network).
I see what you're saying, though could you not say the same thing about the browser's internet cache? That copies the file from it's original server, to the user's local machine in order to display it efficiently.
But the AI is (supposedly) not making a copy of your painting. It is ingesting it, and adjusting it's internal "model of what a good painting looks like" to accommodate the information it gleaned from your work. This seems more similar to what a human might do, when they draw inspiration from another's work. The question is - to what extent does the exact image of your painting remain within the AI's data matrices? That, no-one knows for sure.
Presumably AI trainers aren't hacking into Amazon's servers to steal their copyrighted ebook files. All the data they use are publicly available to view by the AI, just as a human might view them. So I don't think your distinction is accurate. It think the question is to what degree are AI systems "inspired" by the content they are trained on, versus merely regurgitating it - to be honest, it's hard to draw a line between the two for people's creative work, never mind those of machines.
Bear in mind Time magazine is presumably not the greatest fan of large language models, since they'll be churning out Time magazine articles by the 100s, given another year of development.
Imagine an alternative ReLU which had a narrow curved section, to smooth out the discontinuity. Now it has a continuous derivative, but that gradient is still zero for values < 0. This flat region is the cause of dead neurons, because backprop multiplies the propagated error by the gradient, to update the weight - and if the gradient is 0, the result of the multiplication is 0 , and the neuron's weights do not get adjusted.
This already happens to an extent in existing ML pipelines. The ReLU activation function, is discontinuous in its derivative, and is one of the most widely used functions in neural networks. Its derivative looks like this.
if(i<0) return 0;
else return 1;
Now ReLU is continuous itself (as well as being monotonic) so it still cooperates relatively well with gradient descent algorithms. I think this is where the problem lies - not with differentiability itself, but with gradient descent not working due to the highly non-convex search space that such general programming constructs will produce.
Ok, my knowledge of the Fourier transform is limited to the discrete version, but I might as well add my take into the mix. :-) I've always thought of the Fourier transform as a series of projections (as in high dimensional dot products). You dot product your signal with a set of sine and cosine waves, which are selected to form an orthogonal basis over the domain of the signal. So it's just a basis transformation. The phase of the wave is captured by the fact that a sine and a cosine of the same frequency, when added together, will always remain a sine function, but with a different phase, depending on the relative amounts of sine and cosine that are added together.