• zd9@lemmy.world
    link
    fedilink
    English
    arrow-up
    1
    arrow-down
    1
    ·
    2 hours ago

    So do you feel this hatred towards Monte Carlo sampling methods, or Gaussian Mixture Models, or Finite Element Method solvers? It’s all just math and it is being applied towards both how to grow crops better and how to make bombs. Seems pretty naive.

    • BenevolentOne@infosec.pub
      link
      fedilink
      English
      arrow-up
      1
      ·
      40 minutes ago

      You know what all those methods have in common? FUCKING evaluation of smooth continuous functions based on a limited number of samples.

      REAL MEN WRITE REAL PROOFS. They don’t use God damned computational methods which completely IGNORE non-converging regions.

      I used opus to generate this lean-verifiable proof that you in particular are full of shit!

      import Mathlib
      open Real
      
      noncomputable def f (x : ℝ) : ℝ := sin* x) * exp (-x^2)
      
      lemma f_smooth : ContDiff ℝ ⊤ f :=
        (contDiff_sin.comp (contDiff_const.mul contDiff_id)).mul
          (contDiff_exp.comp (contDiff_id.pow 2).neg)
      
      lemma f_zero_on_ints : ∀ n : ℤ, f n = 0 := by
        intro n
        show sin* (n : ℝ)) * exp (-((n : ℝ))^2) = 0
        rw [mul_comm π (n : ℝ), sin_int_mul_pi, zero_mul]
      
      lemma f_ne_zero : f ≠ 0 := fun h => by
        have h₁ : f (1/2) = 0 := congrFun h (1/2)
        have h₂ : f (1/2) = exp (-(1/2)^2) := by
          show sin* (1/2)) * exp (-(1/2)^2) = exp (-(1/2)^2)
          rw [show π * (1/2) = π/2 from by ring, sin_pi_div_two, one_mul]
        exact (exp_pos _).ne' (h₂ ▸ h₁)
      
      theorem sampling_is_a_lie :
          ∃ f : ℝ → ℝ,
            ContDiff ℝ ⊤ f ∧
            (∀ n : ℤ, f n = 0) ∧
            f ≠ 0 :=
        ⟨f, f_smooth, f_zero_on_ints, f_ne_zero⟩