Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

https://en.wiktionary.org/wiki/morg

Morg doesn't seem to be a word in English (though it is in Irish!), but it sounds like it should be.

This is one aspect of AI I will miss, if we ever figure out how to make it go away. The delightful chaos. It invented a word here, without even meaning to.

For example, I vibe coded a QWOP clone the other day, and instead of working human legs, it gave me helicopter legs. You can't walk, but if you mash the keyboard, your legs function as a helicopter and you can fly through the sky.

That obviously wasn't intentional! But it was wonderful. I fear that in a few years, AI will be good enough to give me legs that don't fly like a helicopter. I think we will have lost something special at that point.

When I program manually, I am very good at programming bugs. If I'm trying to make something reliable, that's terrible. But if I'm trying to make a computer do something nobody even realized it can do... making it do things you weren't expecting is the only reliable way to do that.

So I've been working on a way to reintroduce bugs mechanically, by mutating the AST. The fundamental idea is sound -- most of my bugs come from "stuff I obviously meant to type, but didn't" -- but it needs a bit more work. Right now it just produces nonsense even I wouldn't come up with :)

I currently have "mess up the file". The next 2 phases would be "in a way so that it still compiles", and "in a way so that it doesn't (immediately) crash at runtime", (since the whole point is "it still runs, but it does something weird!"). More research needed :)



Nothing delightful of funny here mate. Imagine being a junior just starting out reading that shitty piece of content, trying to make sense of it all?


> Morg doesn't seem to be a word in English (though it is in Irish!), but it sounds like it should be.

Maybe because English also has 'morgue'.


I appreciate humorous outcomes but not when I’m trying to solve concrete task. I’m sure an LLM that is designed to introduce a little chaos is not hard to make. All I know is I won’t miss the weird and incorrect output if they ever get more consistent.


The real question is can we get consistency without mode collapse? Are they orthogonal, or necessarily opposed?

Or to put it more bluntly.. can we get correctness without cringe ;)

I think it could be done, to a degree, with current systems, but it would be more expensive. You'd increase the temperature, and then you'd do more runs. And you could do that iteratively... re-generate each paragraph a few times, take the best of N. So you end up with interesting output, which still meets some threshold of quality.

Actually that doesn't sound too hard to slap together right now...




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: