DeepSeek

The Americans seem to be experiencing their second “Sputnik” in history — Nvidia shares fell by 10%. And the Europeans their first — ASML fell by 10%. Despite the sanctions imposed on China regarding chip supplies, DeepSeek managed to train their model while spending several dozen times less money. I have long been puzzled by the level of confidence that there will be no more qualitative growth and everything needs to be thrown with money. But I am not in the industry, and perhaps it just seemed that way from the outside. Well, now we have practical proof that increasing efficiency is still possible. And I hope that work in this direction will continue, as competition among models is better than a megacorporation that owns everything. It is especially pleasing that the model is open, and its distilled versions can be run even on consumer hardware.