3 ms·
Not only double-check, but probably triple- or quadruple-. There's too much chance of hallucination to trust one or two instances; I think this could be done, b
by flenserboy 3y ago
Not only double-check, but probably triple- or quadruple-. There's too much chance of hallucination to trust one or two instances; I think this could be done, but would require a number of disconnected instances doing the translations, with the resulting documents compared with each other to see a) which ones are more prone to making things up, b) if they hallucinate in similar ways, and c) where they agree, which would allow further study to find out what may have been in their training that allowed similar results to be output.
- sp332 3y agoI would try round-tripping it at least. But I would have to know a little more Akkadian to tell if it's coming up with synonyms or completely different words.
- earthboundkid 3y agoLast week, a story in praise of ChatGPT ended with a haiku that has the wrong syllable count. Why check when it sounds good?
- gattilorenz 3y agoDo you have a link?
- earthboundkid 3y agohttps://nadh.in/blog/this-time-it-feels-different/ https://nadh.in/blog/this-time-it-feels-different/ No shade on the author. It just seems like an apt metaphor for how we get blinded by LLMs.
- sp332 3y agoChatGPT is the only LLM I have seen that can count syllables at all. It seems to be a gap in the way most of them are trained, since they only get text and never “hear” words pronounced.