DeepSeek is a whale shark, not a Black Swan

I'm not surprised by the launch of Deepseek. I'm surprised by all the intelligent people out there who are surprised about the launch of Deepseek. Many an investor has called Deep Seek a Black Swan over the last few weeks.

The transformer technology is well-known. It's open. Altman's empire was built on a perceived oligopoly, where OpenAI as perceived market leader held the throne. Strategically, it was only a matter of time until some company somewhere launched an approach to LLM which is much more efficient than OpenAI's tech stack. This is pure logic. And then we get this politically designed "US against China" artificial flavouring on top. Let's be honest with ourselves; the environmental footprint of OpenAI&Friends is extremely problematic, and the computing costs are way beyond sustainable. It is on high time to upgrade our approach to language processing to a more intelligent level.

Is Deepseek the new OpenAI? Probably not. But that doesn't matter. They have already made a difference. My bet is on more of these unknown challengers popping up around the world, and very likely outside the US.

Tech stocks are falling. Does that mean that the companies are doomed? Nope, it simply means that hot-headed investors are getting a reminder of how the real world works. Hockey-stick shaped dreams are still dreams, even if they have a real-world price tag to them.

Moving ahead, LLMs will primarily be interesting as foundations for solutions to real problems. The value will lie in the solutions to the real problems, not the LLMs as such. Some of the LLMs will be open source, some proprietary. Future LLM architecture will enable quick plug in/plug out of LLMs, making lock-in of enterprises to one specific model unpalatable. In addition, future architecture will both enable and encourage combinations of L(arge)LMs and S(mall)LMs for specific purposes The competitive edge will not be the model itself but the services and infrastructure around them.