Part 6/11:
The focus of the media and investors has predominantly centered on Deep Seek R1. However, the real technological advancement was evident even before R1’s introduction—specifically with Deep Seek V3, which launched a month earlier. While R1 drew attention, V3 showcased significant efficiency improvements, closely rivaling US models in performance while remaining cost-effective.
Dario Amod emphasizes that Deep Seek V3's innovation stemmed primarily from better memory management and an optimization strategy known as the Mixture of Experts (MoE). These advancements allowed Deep Seek to handle larger contexts more effectively and enhanced efficiency in computation.