DeepSeek R1 may be tipping the scale for the future of sovereign AI usage.
At a time where better models necessitate more and more compute thereby centralizing the production and the capabilities of LLMβs R1 has contradicted this assumption and shipped a model that rivals the best in the world with only around 5% of the compute AND they open sourced the model and the training process.
I want a future where I can run the best models in the world in my garage, this seems to be a step in that direction.
π.pdf