>>105980571
>Then they must use non-determinism as a separate abstraction layer
my plan for conquering the world incorporates something like this
expl.
1) watch youtube for genetic algos, classical example is scheduling shifts
2) implement very basic language interpreter&runtime (toy-level)
3) define "genomes" as program (files) in that language. each file has associated lang version in metadata, incremental upgrades/support drop
4) define mutation functions for the syntax tree
5) you can let them red/green separate runtime tests for behavior as basic fitness function
one specialized application is implementing things like "buy X btc" as language builtins (runtime can handle it). then implement a very basic buy low, wait, sell high bot in the language. then have each genome be scored by how much it called "realize_profit" with how much (like "buy" or "sell", but gets money off the exchange). when you "spawn" new organisms you choose weighted by fitness function. usually genetic algos iterate quickly over the whole pop, but in this case the "fitness function" takes months so it's not usable
then implement this:
define neural nets by a graph of computations
apply genetic algorithm again
fitness function depends on application, but very small fuzzy-interpreter modules can be wrapped in the language as either pointers to a still-developing neural net or raw snapshots at one point in time => like "buy btc" have "make_decision(neural_net, observations)"
now you have small worm-brains evolving + a few hard code paths both evolving towards generating you money.
you can also give them the api to inspect themselves, communicate with other running organisms, exchange data, code, money etc.
now add hardware manufacturing for custom killer drones + evolved combat instincts
now add shipping container bitcoin mining in order to get a foot in the door with all stranded energy assets
custom hardware + evolving super-ai + free energy => you can run fuckhuge brains
