Deepseek is so hit and miss right now. That probably speaks to the data it was trained on more than anything. It can write essays better than ChatGPT, but some other things are seriously lacking. I tested it on some haiku and it translated them entirely incorrectly, repeatedly changed the Japanese kanji into the Chinese forms even when I specifically asked it not to do that, got the history of them wrong, the cultural bits wrong, etc. So yeah, major fail. I'm sure it will get a lot better, especially considering it's open source, but right now it's very hit or miss.
You are viewing a single comment's thread from:
I guess it depends on what you're asking it to do. Jamie asked it about a physics problem and it did a remarkable job. So far it hasn't frustrated me as much as chat gippity but I can see how it might do a hash job of haiku translations!