Zhihu
Chinese Quora-like Q&A platform
Episodes
-
#4505: What DeepSeek's Training Data Reveals About Model VoiceWhy Chinese AI models trained on 60% English still produce dialogue that feels distinctively different. -
#2426: Why DeepSeek V4's Prose Feels More Vivid Than Claude or GPTA million-token context window at 2% the KV-cache cost — and prose that actually breathes. Here's what makes V4 different.