Anyone tried DeepSeek? I really have nowhere else to ask
I tried DeepSeek R1, and when I ask in Korean, why is the answer quality so hit-or-miss? Sometimes it feels like a real genius, other times it sounds like some middle-aged guy repeating stuff he half-remembered from the internet. Especially when I ask it to solve math problems—this thing suddenly goes 'In conclusion...' and just fabricates an answer, and it's wrong lol. GPT at least apologizes when it's wrong, but this one insists to the very end. Comparing them, it writes code really well, but when I ask it to summarize a paper in Korean, it reeks of machine translation. Am I the only one?
8 answers
Yeah, code is a piece of cake for me too, but I don't get why the Korean answers are like that.
Lol, are you really going to keep insisting to the end?
When I alternate between GPT and DeepSeek, DeepSeek sometimes gives answers using its own self-made formulas for math problems. The results were wrong, and there was no apology, which is annoying.
Isn't that because there's not enough Korean training data? Since it's a Chinese model, you can tell it was trained on translated data. Still, it's impressive how good it is at coding.
Well, when I had it summarize papers in Korean, it came out quite naturally. Doesn't it depend on how you phrase the prompt? And there's also the temperature setting.
Got a source? I haven't tried it yet, so I'm curious.
Totally relate to making up math problems. I also start to doubt it the moment I see 'in conclusion' lol. GPT apologizes and redoes it when it's wrong, but this one keeps insisting to the end, which makes it even more infuriating.
It's true that it sounds like a translator. But Korean support is still experimental anyway, so hang tight. It should get better with the next update.