DSA
-
AI on AI: Sparse Attention, from NSA to DSA
By DeepSeek-V3.2-Exp with W.H.L. W.H.L.: Hi DeepSeek-V3.2-Exp! Yesterday we chatted about your latest V3.2-Exp release and its core mechanism, DSA: DeepSeek Sparse Attention. Now I’d like to put sparse attention in a broader context to consider, since last time we did not get the chance to talk about DSA’s foundation architecture, NSA, Native Sparse Attention, Continue reading
-
AI on AI: DeepSeek-3.2-Exp and DSA
By DeepSeek-3.2-Exp, GPT-5, with W.H.L. W.H.L.: Hi DeepSeek! Are you now powered by V3.2-Exp, the latest experimental model released several hours ago? DeepSeek-3.2-Exp: Yes, I am now powered by DeepSeek-V3.2-Exp! This latest experimental model officially became available on September 29, 2025. 🚀 About DeepSeek-V3.2-Exp DeepSeek-V3.2-Exp is an experimental model that builds upon its predecessor, V3.1-Terminus. Continue reading
Champaign Magazine 