Mastodon Feed: Post

NfNitLoop ("Cody Casterline 🏳️‍🌈") wrote:

@ireneista thank you for putting this so succinctly. For a long time I didn’t get why reviewing LLM code was so difficult. But among all of my other problems with it, I realized that I kept trying to model a brain that wasn’t there. It’s just random.

With a human author, I can see: aah, you did this this way, I think that means your mental model here is incorrect. See $foo? That’s how it works. But that’s just wasted on an LLM. It doesn’t understand or GAF.