The importance of AI to me is not AI (though I do find it interesting to say the least), it is what it signifies about how our brains work that is really important. It is the randomness. It is our uniqueness. It is how we communicate and yes even feel.
To create images open a browser and enter the https://www.bing.com/create. Then enter a descriptive phrase of an image you would like to see.
All images on this site unless marked otherwise are not copyrighted and cannot be copyrighted since they are machine generated.
All images on this site were generated by AI image creator using a complex model. Each image is unique. Each image is random and cannot be recreated exactly.
This blog is an example of how AI can be used to add images and text to express ideas, and concepts even though AI is still in kindergarten.
Image model
Images created by AI use a diffusion model defined by:
Where T is on the order of a 1000.
So take a 1000 random matrices each 1024×1024 pixels and multiply them together. Then compute the inverse and generate a new random matrix and then multiply by the inverse to create each image.
Those 1000 random matrices are not just any random matrices. They are carefully constructed to destroy the training images by adding small amount of random noise at each step. When finished the training data will just be a random noise (think tv static noise)
Where to from Here
The above model is a 2D model, a flatten view of a 3D and a slice of a 4D world. While simply expanding a 2D model into a 3D model is conceptually trivial, the associated cubic growth in memory and compute complexity makes this infeasible. Added to that complexity is the lack of 3D models for training such models. There are billions of 2D images and by comparison zero 3D images.
The order of magnitude of 3D matrix operations depends on the size of the matrix. For example, if you have a 3D matrix with dimensions 100x100x100, then the order of magnitude of operations would be O(n^3) or O(10^6) operations. If you have a 3D matrix with dimensions 1000x1000x1000, then the order of magnitude of operations would be O(n^3) or O(10^9) operations⁵.
Source: Conversation with Bing, 5/29/2023
(1) Matrix multiplication – Wikipedia. https://en.wikipedia.org/wiki/Matrix_multiplication.
(2) Sort array elements – MATLAB sort – MathWorks. https://www.mathworks.com/help/matlab/ref/sort.html.
(3) Order of Magnitude Calculator. https://www.omnicalculator.com/math/order-of-magnitude.
(4) Vector Multiplication – The Physics Hypertextbook. https://physics.info/vector-multiplication/.
(5) Cross products (article) | Khan Academy. https://www.khanacademy.org/math/multivariable-calculus/thinking-about-multivariable-function/x786f2022:vectors-and-matrices/a/cross-products-mvc.
(6) Order of operations for matrix algebra – Mathematics Stack Exchange. https://math.stackexchange.com/questions/1669280/order-of-operations-for-matrix-algebra.
It is likely that other more computationally friendly algos will be needed.