Hill Space: Neural nets that do perfect arithmetic (to 10⁻¹⁶ precision)

“Hill Space” presents a differentiable system that performs perfect arithmetic and trigonometric functions using discrete neural network weights. No approximations, no external math libraries—just learned neural parameters yielding 64-bit precision. The author demonstrates how these “hill functions” can model addition, multiplication, sine, and more through visual, step-by-step convergence.