Hacker News new | ask | show | jobs
by lifthrasiir 1517 days ago
It is always interesting to see how language models perform for languages not native to them. My trial with Korean (also agglutinative) was amusing:

[Prompt, in Korean:] In order to do machine learning in Hangul [end of prompt; GPT-J still continuing in Korean:] do you think it's needed right now?

[GPT-J switching to English:] English: When the world's about to explode, it's time to learn Korean. If you learn Korean it's time to learn the Japanese language, like the language of Trump. [snip]

2 comments

Freaky. You ever talk to someone right after surgery when they're still high on the anaesthesia and they don't remember anything they're saying? Like, every few words together make sense but as a whole the thing is nonsense. The truly creepy part of the experience of talking to someone like that - similar to this AI - is that they mean to say something a person wouldn't normally say.

When my dad had brain surgery and woke up, I was in the room. The first thing the nurse asked was "who's the president?" And my dad (who actually voted for Trump, I think) couldn't think of Trump's name but he yelled out "He lies like a rug!"

After the second time the nurse realized he did know who the president was.

[GPT-J switching to English:] English: When the world's about to explode, it's time to learn Korean.

This is worthy to be on a T-shirt. Especially these days.