Cyper Bunk
@travisfw building a generative AI model from scratch is not a single copy/paste type thing, and still you're going to use something like pytorch or tensorflow unless you really want to build all the math from scratch too. Even 20 years ago we used prebuilt libraries for the matrix math.
Jul 30, 2023 at 2:54amWeb
Show threadCyper BunkJul 30, 2023
@travisfw and then data collection, cleaning was separate from model training and separate from executing predictions against a model
Show threadTravis F WJul 30, 2023
@bageler I refuse to accept this. There must be a simple example that demonstrates principles, however useless. "Hello world" is useless except you learn to compile code, and write to the console.
Show threadbarrkelJul 30, 2023
@travisfw @bageler it would just be a whole load of loops over arrays multiplying and adding numbers. The structure of these systems is encoded in the weighting of terms in sums, billions of weights, and the programs that evaluate them are more like CPU emulators than anything structurally similar to the resulting computation.
Show threadyarrrivJul 31, 2023
@travisfw @bageler but for most people downloading pytorch is the way they will interact with neutral networks, no? Much in the way that writing Hello World in C abstracts away a lot of assembly underwriting it
Show threadTravis F WJul 31, 2023
@yarrriv @bageler I can't argue with that. I suppose AI isn't yet taught alongside sort algorithms and data structures. (Which are also rarely to never written out by hand, in practice.)