DeepSeek's Open Source Disruption

Ten thousand kicks once will never match one kick practiced ten thousand times. As Bruce Lee famously taught, this martial arts wisdom captures the essence of mastery through deliberate repetition. When we repeat an action with intention - whether it's perfecting your bicycle kick or training a brand new model through reinforcement learning - each iteration builds upon the last, bringing us closer to mastery.

Speaking of deliberate practice and AI advancement, DeepSeek made waves this week with their r3 model. While there are various facts, opinions, and ongoing debates surrounding this development, here's my perspective: OpenAI maintains a significant lead in the field and continues to raise substantial funding, driven by their belief that AGI (Artificial General Intelligence) is within reach. In contrast, DeepSeek has taken a different approach by training on existing research and offering open-source access to their model weights - a move that could democratize AI development.

This might be the end of more money, more models era and the beginning of a new era where the floodgates have opened for open building AI in public. I think this is a huge shift in the narrative and if you have not been paying attention, here are a few links to catch up with.

Deepseek's open-source approach to AI development challenges traditional models, proving that transparency and community collaboration can rival well-funded competitors. Their success demonstrates a potential shift in how AI technology evolves and is distributed.

https://medium.com/@anixlynch/tech-wise-how-deepseek-beat-openai-with-just-5-of-their-budget-6aa30bbe8677

A comprehensive analysis of how DeepSeek's open-source AI models are challenging OpenAI's dominance, exploring the implications for the future of AI development and market competition.

https://www.vox.com/technology/397330/deepseek-openai-chatgpt-gemini-nvidia-china

An in-depth technical analysis exploring DeepSeek's V3 and R1 architectures, performance metrics, and real-world applications.

https://www.youtube.com/watch?v=J5xZIw7G1NI

I’ll be speaking at State of Open Con 2025 next week about Open Source Metrics.

Details here: https://stateofopencon.com/schedule-2025/

Regardless of who wins the AI wars, I'm excited to build with this technology and am working toward becoming an AI engineer. I received curriculum from a friend who's a UC Berkeley alumnus and will begin developing my first AI side project in February. I'll share more details about that next week, so stay tuned.

For now, the one place to run Deepseek v3 without giving your data to the People's Republic of China is Together.ai. You can test it directly on Huggingface for free, you will to have a HF log in to test a model.

https://huggingface.co/deepseek-ai/DeepSeek-V3

If you find the contents interesting, feel free to comment or share with a friend. Future iterations will focus on me building with AI and reporting my findings.

I'm working on relaunching the Secret Sauce podcast and will be reaching out to potential guests soon. Open Source Ready already has a great guest backlog, but we're always looking for more. If you know any thought leaders or founders who'd like to discuss AI and open source, please send them my way or forward this email.