Surfing the Singularity - The World is Not Flat
As Bill Gates recalls in his recent book-bumping interview with the Wall Street Journal, in the early innocent days of Microsoft he and his co-founder Paul Allen didn't believe in having an office in Washington, D.C.[1] They were soon to learn that was a mistake.[2] Compare and contrast with the scene in the Capital Rotunda last week for the inauguration of the new populist administration - Microsoft, Amazon, Facebook, Apple, Google, TikTok, and of course Tesla, all represented by their CEOs.[3] Microsoft's market capitalization as of this writing is now greater than the GDP of France.[4] Elon Musk's personal wealth is on par with the GDP of Denmark. Meta's platforms reach an estimated 40% of the world's population.
| Apple ad from yesteryear. We're now long past 1984. |
Consider these other inconvenient truths about global technology: that NVIDIA does not make the GPUs it designs, that most are manufactured by TSMC in Taiwan, which is about as far away from China as Cuba is from Florida. Software talent is globally distributed, and prices vary widely. There are some very good schools in some of these relatively inexpensive places - in the 2024 edition of the ACM student programming contest, MIT placed the highest among US teams, in 11th place.[5]
| Software salaries 2023, by country, exchange rate normalized.[6] |
AI programs and quantum computing initiatives are increasingly becoming nationalized as a strategic imperative. It is assumed that the country which is first to Artificial General Intelligence (AGI) will be the first to be able to use it to take control of the world. This fear is fueling an arms race in AI architectures, and the chips and the energy stations which power them. But is this a rational fear?
Unlearn What You Have Learned
DeepSeek, with its deja vu inducing name and similarly eerily similar whale icon, is a new AI chatbot model wholly owned out of Hangzhou, China. And its #1 on the Apple app store, with a bullet. And yes, its quite up front that it tracks your data. There are several notable features and claims about this model: that is was trained in a fraction of the typical cost and time on a fraction of the typical hardware. That it performs about as well the OpenAI model released last month, maybe not as well on some things like straight math, but perhaps better at general writing, maybe with a bit more "personality". It shows its work - how it arrived at the answer its providing to the chat prompt. And, for the kicker from the totalitarian state, its open source up on Hugging Face.[7]
Shots fired. NVIDIA shares tumbled for a loss of $600B in one day in response, the largest single day loss in history. But what I'd like to know is, if this is the open source model, what's the real one like? Recent Federal governments have tried to enforce policies limiting technology exports to China, but in truth they've long had their own development programs - China has not participated in standardized HPC metric sharing and supercomputer ranking since 2017.
Red Flag on the Track
Along with NVIDIA, power generation stocks were also down hard - GE Vernova down almost 20%. This does not mean the recent trend of tech companies buying nuclear power plants won't continue, or that we won't continue to hear of existing power stations giving data centers a direct hard wire bypassing the municipal grid. But clearly this open ended hunger for power - watts and GPU cycles - is not sustainable. And DeepSeek exposes that bare.
But make no mistake - this race is not over, its just warming up. OpenAI with their new Operators product currently defines AGI as a gaggle of collaborating AI agents, each with its own unique set of capabilities and goals. NVIDIA CEO Jansen Huang does his part in driving the GPU-dependent AI hype cycle by saying IT departments will become the new HR departments - for AI employees.[9] Goldman Sachs is telling clients to expect AI employees this year.[10] Cost avoidance will be a major driver.[11] And why not, when the same major technology companies report seemly amazing results using AI for software development? Google saves 50% on code migration time with AI! It gives *me* FOMO! [12]
The CEO of Anthropic predicts that by 2027 AI will be generally better than humans at almost everything.[13] Well, at some things maybe better than others. Turns out, what's the number one occupation we expect to be replaced by AI? Why, AI engineers and data scientists.
| Job skills impacted by generative AI, ranked. Maybe I should have been a plumber.[14] |
This is vast uncharted territory for companies, especially for those of size, wedded to their legacy political structures and being either un-nimble or worse, fragile. People are not machines. The mistakes AI agents make are not of the same kind made by humans - current generative AI models are intentionally designed to make stuff up, to not just say "I don't know".[15] As a manager, you'll get no benefit of human insights into the truth such as from body language, though you may get to understand the "personality" and general performance characteristics of your AI employees over time. That is, until the managers are replaced by AI. But until then, what does your management interface look like? Is it perhaps similar to the IDE for a senior software engineer who manages a team of AI coders? In the AI-laced future, there will still be a place for HCI/UX designers.
Chef of the Future
Finally, the November 2024 report from the National Academies on the "future of work" says "its impossible to predict exactly the nature of the coming changes in AI and all their effects on the economy and society".[16] This includes how it changes the nature of various jobs, or outright eliminates them. Continuing education will be key to a resilient workforce, and it turns out, AI might even play a role in that.
As shown in Washington last week, there are new business trends being driven by a rejuvenated alliance between big technology and big government, and it is therefore time for astute technology and business leaders to pay attention to both.[17] For example, tomorrow (January 30, 2025) OpenAI is holding a previously scheduled closed door meeting in Washington regarding its own current agentic technology innovations, and what they potentially imply for the US people and its government. I imagine the topic of DeepSeek will now disrupt the meeting agenda, somewhat.
At minimum, its a topic for a future blog. Regards. - andy