Recently saw this gem (have not actually verified, because it's funnier to just believe it to be true).

There’s probably something in the chat session preceding this, but I’m pretty sure it is legit otherwise. The basic mechanism of ChatGPT is to use its training data to predict next word. Like a Markov chain on steroids. A perfect bullshit generator.

They caution on the website that it can (and does) generate perfectly plausible incorrect answers... this should be no surprise 🙂

I'm not surprised other than that it has some sort of ”idea” what Calva even is.

I wonder if the Clojurian's log is included in its training data. I think that maybe it hasn't, and that if it had it could get more helpful with Clojure things.

Every question it is asked is part of its training data too. So if anyone asked it about Calva and it guessed wrong, they might try to "teach" it by guiding it to an answer.

The more Clojure-related questions people ask it, the more plausible its wrong answers will sound 🙂

Yeah, just tried that 1000 larger than 1062 and got the same reply

From what I've used it for, as an assistant essentially for coding, i've found it very impressive so far

It calls itself Assistant for a reason. 😃 https://blog.agical.se/en/posts/chatgpt--are-you-holding-it-wrong-x-/

yeah, i suck at sql and was scratching my head getting a query working, asked GPT and it just spat out the working query

I'm kinda using like an actual good version of stackoverflow, without judgemental assholes telling me my question is dumb / off topic / a duplicate.

I didn't know about Expectations. Found a clojure.test compatible version: https://github.com/clojure-expectations/clojure-test