Just learned about Bun and Deno, Javascript runners/runtimes/package managers.
Bun in particular has a lot of Zig and C++ explaining the speed that its website boasts. On my CPU-only machine it’s great. While Deno uses Rust and Javascript/[Typescript](https://www.typescriptlang.org/.
Found valuable resources for improving MNIST such as Techniques to Improve Our MNIST Accuracy Without Using CNNs (99.46% accuracy) as well as a repository of articles and their respective code and a website with a warning:
Kaggle works on a system of honesty, there are ways to fake the results of your ML algorithm, and there are lots of fake results on the platform. If you look at the MNIST leader-board on Kaggle, you’ll see a whole page of people who claim to have scored 100% accuracy, which is impossible. If you look at their code, all they are doing is passing through the test data as model out1puts without any machine learning.
Also found MNIST 99.6 % Accuracy by accident.
Thanks to this guide for the Markdown instructions.