Again, I am just experimenting but the way I plan to introduce this is through precisely controlled KJ buckets. Multiple intensity “buckets” like 400-500w, 360-399w, 320-359w, and 275-319 (which is just threshold region).
Take each block and grow them at a ramp rate per day in KJ. For example, add 1kj per day into 320-360 range, 2Kj per day into 275-319 range.
Now these aren’t traditional methods, not at all. Im not claiming its correct. But the issue is that from a first principles point of view, the current methods in training all converge around certain assumptions. I’m trying to break those fundamental assumptions.
My inspiration for this plan is the work in Simulacra and Simulation as well as DARPA’s Gamebreaker
I also realize a lot of readers of this post wont grasp these ideas and thats ok. So they’ll probably go for ad hominem instead lol.