Itd be cool if we could get something like Schenzen.Io going, but you build the chips from the gates up
Maybe package them into modular units, and connect those at a higher level or abstraction, ad infinitum.

Then add access to virtual LCD output, and other peripherals, or even map output to real hardware, essentially letting you build near bare-metal virtual machines.

Dont know about that last part, but the closest I've seen to the rest is circuit simulator and again, schenzen.

On the machine learning front I figured out I need about ten times as many training samples as validation samples, or vice versa. I'll have to check my notes. Explains why I could get training loss below 2.11

Also, I'm looking at grouping digits, and trying different representations. I'm looking at the hidden variables for primorials to see what that reveals. And I realized because of the amount of configurations and training that I want to do, even a personally built cloud isnt going to be sufficient. I'm gonna have to rent someone else's hardware and run it "in duh cloud."

Any good providers that are ridiculously upfront for beginners to get started with? Namely something cheapish.