In the last month, the new Chinese AI, DeepSeek, has disrupted the AI world with its new technology! Or maybe not. For years, the West has claimed that China does not innovate. Instead, Western sources say that China steals technology others invent and implements it on a larger scale. What’s really going on? Is DeepSeek new news or just another Chinese knockoff? Let’s dive in.

At the end of January, DeepSeek announced that it had a better product than the world’s leader, ChatGPT. While it seems quicker and more energy-efficient, is it truly better? A quick glance will show you that it lacks many of the features of ChatGPT.

The interface looks like a straight-up copy of ChatGPT… not a lot of innovation there. ChatGPT spent a lot of time and energy on its voice interface, which DeepSeek does not have. What DeepSeek does have is a far more comprehensive censorship system. Ask about Tiananmen Square. Ask if one million Uighur people have been imprisoned by Xi. Or if it was 15 or 30 million Chinese that died under Chairman Mao. Suddenly, DeepSeek is unable to answer the question.

DeepSeek’s deep censorship is definitely a Chinese innovation. I asked DeepSeek about this, and after a lot of deflections, the answer came down to Chinese law. It is simply illegal to address issues that the Chinese government identifies as illegal. In the most general sense, anything that could embarrass or harm the reputation of the Chinese Communist Party is illegal. And that, dear readers, covers a lot of territory!

Many reviews have mentioned DeepSeek’s censorship. You can add to that the likelihood that it will collect and/or steal personal information and corporate intellectual property. That’s bad enough, but the censorship is likely to distort information, thereby making the results questionable if not misleading. It is a matter of social rather than technical limitations, but it is a limitation nonetheless.

However, the main positive development of DeepSeek is the Chinese invention of a new technique, Distillation. Essentially, you reduce a larger dataset into a smaller, more manageable one. Naturally, the smaller set can be searched more quickly and takes less energy to accomplish. That’s brilliant! You really have to take your hat off to the Chinese for innovating rather than merely copying.

Just kidding. Distillation was developed in 1991 by Jürgen Schmidhuber. Distillation was a refinement of an earlier system called Pruning, developed in 1965 by Alexey Ivakhnenko and Valentin Lapa. Although to be honest, we should give credit to Clifton Keith Hillegass. In 1958, he developed this methodology in the basement of his home in Lincoln, Nebraska. Hillegass was not a scientist, mathematician, or computer developer. But he instinctively understood the value of data compression. You all know his product, and most of you have probably used it: Cliff Notes.

When faced with a massive repository of text (i.e., a novel), many high school and college students failed their test of data retrieval (i.e., the final exam). But Cliff Notes saved their academic careers. The secret behind DeepSeek was the Cliff Notes model… and everything that has been built on it since.

Merely imitating the work of others is a lot easier than the original creation. The executives at DeepSeek said that it was created with completely original code for a mere $6,000,000. Did DeepSeek spend $6 million to figure out that it could not catch up with ChatGPT’s level of innovation? Very likely. I believe that they then decided to focus on the speed of code execution. Why do I believe this?

Once it was clear that Distillation (plus other techniques) was the source of DeepSeek’s “innovation,” others quickly followed. Researchers at Berkeley recreated OpenAI’s reasoning model in 19 hours, for just $450. Stanford and the University of Washington created their own reasoning model in 26 minutes, for under $50. In a 24-hour coding challenge, an AI development group recreated OpenAI’s newest feature (Deep Research). We’re told that it took $6 million, 6 months, and over 200 developers to create DeepSeek. Rather than being an example of the world’s best AI development, it looks like the worst use of $6 million.

What do you think? Is there anything innovative about the DeepSeek system? Is it trustworthy? Would you rely on it? Or what about AI from the rest of the world? Let’s hear from you!