Post

Shoving non-deterministic AI model outputs into constrained execution environments with deterministic feedback works great. I'm less sure it will work for other contexts.