Comments by "Brent Jacobs" (@br3nto) on "Meta drops new LLM based testing" video.

  1. We’re getting so close to not having to code…. They just need to integrate CSPs, LLMs, and whatever the AST manipulation technique is, and boom! AI that could verifiably write code that meets the constraints given to it. Just open an issue, assign it to an AI helper, and in a short period of time you’ll have your new feature. I can’t wait!
    1
  2. 15:36 1326 lines doesn’t scream “unit” to me. Would that test even be understandable? Will it break at the drop of a hat?
    1
  3. I don’t think lines of code covered is a good metric for test. The metric for a good test should be: does it test a requirement, and does it effectively validate that my program meets that requirement.
    1