Aug 13, 2025 10:59 am
Sterfteerst says:
The generative models we call ai are still bound to be linear processes. Mathematic instructions are followed sequentially because that is all our computer hardware can truly do.Meanwhile a human is still non-linear. Keeping in mind several considerations at once and recalling things all over.
There are complexities that our current computer hardware, by the fundamental way it works, cannot replicate in an efficient manner. That sort of computing needs either different hardware, or incredible amounts of processing.
AI models represent data as vectors. A vector is an object with a bunch of numbers. We can think of a DND character's abilities as a vector of ability scores, HP, skills (e.g. [16,12,8,14,11,10,24,0,0,1,2,0,1,...]).
When you stick a bunch of numbers together, you get a matrix. AI models do this to process their data as matrices, as it's then possible to perform mathematical operations on a whole lump of data. It allows processing of information as a gestalt. Going back to our example, we could imagine a matrix which adjusts HP based on CON, and then apply it to a party (where a party is a bunch of vectors jammed together into a matrix).
I don't know if you remember how to perform matrix algebra from school, but if you do, then you might remember that finding numbers in the result matrix can be done in any order (although we tend to work left-right, top-bottom out of cultural habit). This allows matrix operations to be parallelised.
This is where GPUs come into play. GPUs are built to allow massive parallel processing and are optimised for this sort of processing. It's also why Nvidia (a GPU manufacturer) has become a very valuable company on the back of the AI boom - they provide hardware to help make general matrix multiplication parallel and non-linear.
Now, of course, within those myriad parallel streams, there are still sequential operations, but massive parallel processing of matrices is still fundamental to current AI models.
I didn't mean this to come across as an "ackchyually...", but more of an answer to anyone who ever asked their maths teacher, "when am I ever going to need to calculate an eigenvector?"