DeepSeek sent the capital markets into a tailspin this week with the release of their latest model. Remarkably, DeepSeek trained this model at a cost of under $6 million, significantly less than the estimated $100 million spent by US firms on similar models. The innovation was leveraging the training of other Large Language Models (LLMs) to dramatically reduce the amount of compute required to provide the same or better performance than the best existing models. The technical aspects of the technology are beyond the scope of this article.
The scariest element of DeepSeek's apparent breakthrough is that it increases the competitive pressure between the US and China — making it much more likely that AI safety will take a backseat to unrestrained progress. China and US Tech were never on my shortlist of careful actors, but this certainly aggravates the situation.
The nightmare scenario is what the AI experts call an intelligence explosion, where smart AI is able to work recursively to make better versions of itself. One can only imagine the progress that could be achieved if an AI was able to harness a warehouse full of Nvidia's latest chips to make itself smarter.
What if AI was able to do the work of 10,000 people working for a year, but the AI would complete the job in an hour? Obviously, it does not take long for this process to get out of hand. One could imagine the level of intelligence achieved after several thousand iterations. Think ChatGPT version 3000.
Is anybody asking what would be the trigger to pump the brakes on this technology? I hope the answer isn't, "we'll know when we get there."
The Fermi Paradox is the apparent contradiction between the high probability of extraterrestrial life and the lack of evidence or contact with such civilizations. It is named after physicist Enrico Fermi, who famously asked, "Where is everybody?" during a conversation about the likelihood of intelligent life in the summer of 1950 during a casual lunchtime conversation with colleagues at Los Alamos National Laboratory.
Fermi was one of the principal scientists who worked on the Manhattan Project and developed the atomic bomb. He certainly was aware of the abilities of human beings to harness the power of science and saw firsthand the destructive forces that were unleashed.
The Fermi Paradox has several possible solutions. I would encourage anyone interested to read the Wiki article. The scariest solution to the paradox is called the Great Filter, which proposes that every intelligent civilization would undoubtedly follow a similar technical evolution as the laws of physics are the same everywhere. These alien civilizations would eventually discover atomic physics and computer science. At some point, the technology becomes too powerful to control, leading to the end of the civilization.
Let's all hope that the Great Filter is not unrestrained AI.