DeepSeek is a good hack.
It uses Agile programming, parallel processing, and a focus on efficiency to deliver a Large Language Model that can train databases using a tiny fraction of the computing resources needed by OpenAI, Gemini, or Co-Pilot.
Plus, it’s all open source.
DeepSeek is a proof of concept that does for AI developers what the PC did 50 years ago. Back then, computing meant shoving a stack of punch cards through a window, then getting a printout some hours later saying you’d typed something wrong on card three and you wouldn’t be seeing any output. The turnaround on each mistake was a full day. On an Apple II it took me a few seconds to figure out I was no programmer .
The arguments against DeepSeek are political.
Of course it doesn’t know about Tiananmen Square. It was trained on Chinese data. Train it on Fox News and it won’t have heard about Trump’s rape conviction. Of course it’s subject to hallucinations. That’s inherent in all LLMs. They have no physical presence, no concept of the real world. They only know what they’re programmed to know.
DeepSeek is a framework you can use to build any model you want, trained in any way you want, aimed at any target you want. Garbage In, Garbage Out. But clean data in will get high performance out. The database you’re trained on means more than how the database is trained.
In the end there is no such thing as AI Supremacy. There is just the software you’re using and the data it’s trained on.
Elmer FUD
In just two weeks DeepSeek has created a cycle of Fear, Uncertainty and Doubt (FUD) it would have taken Bill Gates months to concoct 40 years ago. Taiwan has banned it and one Senator has proposed jailing people who download it.
This is insane because, again, DeepSeek is open source. You can change it as you like. Its parallel processing concepts, the idea of handing different types of questions to different AI “experts,” make this sort of thing simple.
At the heart of the overreaction is the idea that DeepSeek belongs to China. It doesn’t. Trying to control open source is like trying to hold water in your hands.
However much money it took to build DeepSeek, how many compute hours were used in its development, the fact is it’s out in the world now. Tweaking it will cost a fraction of what it took to build. Its concepts are available to anyone. Right now, some kid in Harare could be writing something that makes DeepSeek look like OpenAI, because he has the code.
We go back, in the end, to the four freedoms that open source grants us. The freedom to use the code, to see the code, to change the code, and to seek profit from those changes. If you want to focus on any political message from all this, focus on that.
