AI's Jarring Wake Up Call

Phase One of the AI Revolution is Complete. Now Things Get Interesting.

Should we really be surprised by the DeepSeek news?

In case you’ve been hiding under a rock the past few days, let me catch you up. DeepSeek is a Chinese AI company that figured out how to build a reasonable LLM AI model that competes with your favorite Western AI model like Meta’s Llama or OpenAI’s Chat GPT 4.o for a fraction of the price.

The company claims it produced its model for the equivalent of $5.8 million, way less than what other companies spent on the increasingly cutthroat war of attrition to create the latest and greatest AI model.

More important than even the cost of development, DeepSeek built the model with a very pedestrian tech stack of chipsets that was considered necessary to build a sophisticated AI model. Apparently DeepSeek has done so by utilizing open source code that anyone can help improve rather than proprietary source code of the entrenched companies in the industry.

The markets woke up Monday to the realization that many more companies might upend the AI leaders AND that the companies that will power building out tech stacks will be far less valuable than was considered just…last Friday. Naturally it was a bloodbath on Monday. The AI chip company Nvidia alone lost 17 percent of its value, around $589 billion in value in a single day.

The market is clearly over-reacting and I’m sure that some of the losses will turn around in the coming days, although a correction to the bubble of AI hype that has run up the value all tech stocks is utterly unsurprising, if not overdue.

But I would suggest that our priorities for the AI Boom have been out of whack for a while. There’s been an arms race on building better, more sophisticated models in a race towards Artificial General Intelligence and far too little on how consumers use these tools.

Put another way, instead of being shocked that a Chinese company figured out a clever way to make a Chatbot agent so much more cost efficiently, I wonder why the hell we’re still talking about Chatbot agents at this point? It’s been two plus years since Chat GPT debuted. Shouldn’t we have moved past the Chatbot model by now, or at least worked towards improving it?

One of the most interesting things about the DeepSeek app is that it shows the consumer its reasoning and thought process when answering a prompt. It’s something none of the other services allow, as they are trying to keep its reasoning proprietary. While I understand the strategic reasons for hiding the logic, it really does make for a maddening the user experience, since it takes trial and error to figure out how to get the service to answer the question you asked correctly. By knowing its reasoning, you cut down on the guesswork to get to the answer you’re looking for with DeepSeek.

So, it took a company that doesn’t have the best tech, that is operating on a much smaller budget, and is in the backwater of AI development to come up with the most meaningful user experience innovation in the past couple of years. If that’s not a wakeup call for the industry, I don’t know what is.