Working with DeepseekR1

If you haven’t heard there is a new model on the block. It didn’t come from Stanford, UW - Seattle, or Harvard. In fact it didn’t even come from North America. It came from overseas and was built in a country which currently is under certain export bans as well. In this post I will cover what is DeepSeekR1, what is new - or not new, and does it shift the balance of AI innovation from the US to overseas.?

First, What is DeepSeekR1?

DeepSeek is a side project of the quantitative trading firm baed in China called High-Flyer. It started out as a side project or research project by the company. The original name for the project was Fire-Flyer.?

High-Flyer had been stockpiling gpu’s for their research projects around trading funds and currencies. The original idea was to create a company that would be called DeepSeek as a tool to build AI models based upon synthetic data.?

It worked.?

The primary purpose of DeepSeek was to build on a long term philosophy and so the main idea was to create a tool for research and not so much for commercialization first. The secondary purpose could be commercial, but since it is now open source you can use it basically for free.?

Knowing this?Cuppa.sh?has already started allowing users to access DeepSeek while writing content. Cuppa allows you to access an LLM via an API key. This makes using Cuppa to write content much cheaper than using something like Chatgpt straight in the browser.?AnythingLLM?is an open source tool that also allows you to use an API to write, rewrite, rephrase your content. And it also works with DeepSeek API keys.?

To learn more about using DeepSeek API keys try this?article?on the DeepSeek website.?

My Experience with DeepSeek

I have only begun to use DeepSeek through an API or through the terminal at this point personally I am not impressed. The writing still has that bland AI readability that overuses words such as ‘workflow’, ‘crucial’, ‘In today’s fast paced world’ and so forth. The writing itself on top of that is?not superior to what I have found with Chatgpt?or Claude.?

In fact I have had to take DeepSeek output and run it through Chatgpt in order to get something useful out of it. I do not find it superior to current models from Anthropic or OpenAI or even Gemini. Those are my goto models when I need/want help with my writing or reserach. Asking a question of Chatgpt almost always gives me a useful answer.?

Ending Thoughts

There are other problems with DeepSeek such as privacy. Their terms of use state they have full right to any output coming from their model use. Some are starting to tell others?about this too.?

Also, DeepSeek is not reinventing the wheel. It was built upon data and output coming from Chatgpt. Though I think Open Source is the future of AI models this is not the model that proves that yet. As has been?pointed out in this post copying an LLM is not that complicated for any with a bit of knowledge. It is merely trivial to do so.?

Since this is essentially a copy, if you will, of already in use models does it end American dominance in this space? Not quite. Does the development of AI models need to be more democratized? Yes. Is Open Source the way to go? Likely, as copying a fully researched and betted AI model is not as complicated as we had thought it to be.?