Hostinger Horizon: https://www.hostinger.com/calebwritescode
Coupon Code: calebwritescode
China's most unorthodox AI Lab, DeepSeek, released a paper called mHC, or manifold-constrained Hyper-Connections which is creating a lot of discussions around how this could influence LLM architecture of the future.
As we look at how DeepSeek and ByteDance is growing their AI dominance, we have to get technical in this video to understand where Chinese model optimizations are occurring.
#ai #deepseek #largelanguagemodels #deeplearning #artificialintelligence
Chapters
00:00 Intro
00:56 Trend
01:41 Adding Intelligence
03:25 ResNet
04:17 Sponsor: Hostinger
05:16 Recap
05:57 LayerNorm
06:44 Post vs Pre-LN
08:08 Hyper-Connections
09:06 mHC
10:45 Conclusion
