Deep(Seek) Thoughts, a current thing

This post is off the path of where I think the is the puck going. This is about where it is right now, as of when I push publish.

The China vs US geopolitical US vs China are very real and part of anyone's thinking. But what what else is at play here, and how might we look at DeepSeek's training approach and the possibility they used o1 or models from OpenAI?

I don't have much of a horse in this race, this is just an approach that might be helpful to be clearer with yourself about the position you're taking.

Let's break this down into legal, ethical and moral concerns.

Legal

The legal issues are a mix of statute, regulation, case law and contracts which this includes terms of use.

The closest we have is the Copyright Office guidance released earlier this week.

“After considering the extensive public comments and the current state of technological development, our conclusions turn on the centrality of human creativity to copyright,” said Shira Perlmutter, Register of Copyrights and Director of the U.S. Copyright Office. “Where that creativity is expressed through the use of AI systems, it continues to enjoy protection. Extending protection to material whose expressive elements are determined by a machine, however, would undermine rather than further the constitutional goals of copyright.”

Unfortunately we will have to wait until Part 3 of the Artificial Intelligence Report for guidance on training.

While there aren't directly applicable statutes or regulations for AI, the DMCA and its Anti-Circumvention provisions may be relevant. If DeepSeek reverse engineered o1's CoTs for use in reinforcement learning, they could be in violation.

We should have at least one court decision soon, with the highest profile candidate being the New York Times suit filed against OpenAI and Microsoft. I'll link to NPR, but there are many more accounts.

So the most relevant legal issues are in contract law. Assuming there aren't any side deals, what matters here are the Terms of Use for any services used. Assuming DeepSeek used Common Crawl and similar data sources to train the base model, they're in the clear here.

However if DeepSeek used OpenAI at all they've violated OpenAI's Terms of Use found here. Section 2(e) says "use Output (as defined below) to develop any artificial intelligence models that compete with our products and services.".

The most likely uses would be generating

(1) Synthetic data for the base model or

(2) Chains of Thought for reinforcement learning

Ethically

Ethics is something like "expected practices in a category that may or may not be formalized". So the question is what is the area, and are there expected practices.

If we saw this is in the software category I like the now decades-old Open/Closed source dichotomy where Open decomposes into GPL (free as in speech) and not GPL (free as in beer), versus Closed. Playing it out, Free as in Speech is going to be OK with pretty much anything DeepSeek does. Ask me how I know and I can share a story about Symbolics when my uncle worked there.

Free as in Beer is fuzzier and hard for me to define without another cup of coffee. I'll just say this is the space occupied by people who are not Stallman-level open source advocates on one end of the spectrum and people who are not software patent absolutists.

Specifically for for AI research labs, which is what OpenAI and others sometimes call themselves, I am not aware of a set of ethics. However I do not think this applies, despite the tendency of employees of OpenAI and Anthropic to refer to themselves as researchers. The reality is these are all profit-seeking enterprises. Although this has been a relatively new development for OpenAI on paper, I'd argue it has always been true for them.

Morally

In this newsletter, when I say morals, I mean something "universal". I mention this area to help avoid blind spots and biases, not that I know what the right morality to apply is.

This is where beliefs about property rights and more fundamental humanist and theological and similar beliefs reside. Judeao-Christian frameworks, Rationalism, Effective Altruism, and so on.

In my experience, even if you pick what seems to have a narrow enough lens to be useful, there's usually still plenty of room for disagreement. I'd best to just try to be explicit about what pieces of which moral framework you've pulled into your position.

Conclusion

I think most of the confusion here stems a misapplication of ethics as well as reliance on unarticulated morality.

The ethics here seem to be Open for me but not thee.

• Open for me: I'm an AI lab, so apply an academic or research standard to my inputs
• But not for thee: You're a competitor - fellow market participant or foreign adversary - so your inputs should be commercial or just prohibited.

Hopefully separating legal, ethical, and moral concerns helps see where they overlap and inform, or even contradict each other.

Thanks for reading this far. It was fun to use some old skills from my undergraduate and law school days.

Also, slide into the DMs, add a comment or text me with your thoughts.