@sanityinc @defuneste Yeah, I was active in both the Clojure and TDD worlds at the time, and it was clear Hickey was talking about a different kind of testing than I was doing. The kind I’d been doing in the ‘80s, where testing was not used as a tool for thinking about a problem or the design of a solution.

I had roughly the same attitude as Hickey until I sat down and used TDD on a serious project. I don’t know if he ever has. I’ve been out of Clojure for over a decade (!)

@sanityinc @defuneste It’s sort of an interesting meta-commentary. Both of us thought we could discard a methodology by reasoning about it rather than trying it. I was lucky that @RonJeffries argued me into giving it a try.

Part of a career of reasoned-through conclusions that didn’t stand up against experience. Eventually I figured out that maybe I’m not smart enough to get it right by sustained cogitation, so I needed to adopt an approach more driven by rapid feedback from the world.

@thirstybear @marick @sanityinc @defuneste @RonJeffries

You should feel free to reexamine any belief at any time for reasons you find compelling, of course.

But in areas like this I apply Bayesian probabilistic reasoning, where I take (1) what I already know, and combine it with (2) additional new evidence, to determine (3) how much it makes sense to change my beliefs. With LLMs, it's generally "not much," with all the claims that "Oh! The new models fix EVERYTHING!" when they don't.

…

@thirstybear @marick @sanityinc @defuneste @RonJeffries

Also, I think it reasonable, practical, and really necessary to apply some amount of "the halo effect" and even some "argument from authority," when the accepting input from someone who has *relevant* experience and authority, who has consistently demonstrated competence and correctness in the past.