Zalgorithm

The year of the midwit

My favourite LLM interaction (I’d asked it to check my work):

ComplexRect.scalarDivide() — completely wrong logic #

ComplexRect scalarDivide(double s) {
  double denominator = Math.pow(s, 2);  // ❌ Why square it?
  if (Math.abs(denominator) < 1e-10) {
    return new ComplexRect(Double.POSITIVE_INFINITY, Double.POSITIVE_INFINITY);
  }
  double real = (re * s) / denominator;      // ❌ Multiplying by s then dividing?
  double imaginary = (im * s) / denominator; // ❌ Same issue
  return new ComplexRect(real, imaginary);
}

The issue being that it was the right logic. A scalar is a complex number. For example 17 = 17+0i. My approach was just an inefficient implementation of the divide method for the case of other.im = 0:

  ComplexRect divide(ComplexRect other) {
    double denominator = Math.pow(other.re, 2) + Math.pow(other.im, 2);
    if (Math.abs(denominator) < 1e-10) {
      return new ComplexRect(Double.POSITIVE_INFINITY, Double.POSITIVE_INFINITY);
    }
    double real = (re * other.re + im * other.im) / denominator;
    double imaginary = (im * other.re - re * other.im) / denominator;
    return new ComplexRect(real, imaginary);
  }

The year of the agent #

LLM midwittery aside, I’m jumping on the OpenClaw bandwagon.

Sort of… https://learn.deeplearning.ai/courses/agentic-ai/

Tags: