DeepSeek AI's Cost-Efficiency Breakthrough: Professor Wong Yan on China's Path to AI Market Domination and the Limitations of Current AI

Professor Wong Yan presents a compelling argument that the AI race, particularly within the current generation of Large Language Models (LLMs), is effectively over, with China poised for dominance due to a significant breakthrough by DeepSeek AI. He posits that while many new technologies exist, only AI is capable of becoming a beneficial industry within a reasonable timeframe. The current AI generation, starting around 2012, represents the fourth or fifth iteration of AI, following decades of research with little commercial return until recently. Despite its commercial value, Professor Wong expresses skepticism about a 'killer app' emerging from this generation that would fundamentally leapfrog human civilization, primarily due to two major inherent flaws.

The first major flaw identified is that current AI operates on 'probability elimination' rather than 'active reasoning.' It generates the most likely answer based on vast online data, essentially 'guessing' or 'pretending to understand,' which limits its application in critical areas like medicine or complex, unique problems such as Olympic math. This explains the persistent challenges in areas like self-driving cars. However, Professor Wong notes that current AI is well-suited for tasks like writing papers, image and video generation, and accelerating programmer workflows, all of which hold significant economic value. The second, more fundamental flaw, is the 'computing efficiency frontier,' a mathematical constant where doubling AI accuracy requires a million-fold increase in computing volume, making further scaling prohibitively expensive and energy-intensive, a key driver behind Nvidia's high stock valuation and the demand for massive power infrastructure.

DeepSeek AI's breakthrough, according to Professor Wong, lies in its ability to bypass this computing efficiency frontier. By dividing AI into modulated parts and training them separately, DeepSeek achieves similar performance with roughly 100 times less computing power. This innovation, anticipated by Professor Wong two years prior, represents a significant leap in efficiency, resulting in approximately a 20% accuracy increase for the same computational resources. Crucially, DeepSeek's development and operational costs are only 1% of existing market alternatives, a disruptive change likened to a car achieving 1,500 km on one liter of gas with a $50 upgrade.

This cost reduction fundamentally collapses the entire price structure of the AI industry, rendering previous massive investments in high-performance chips and power infrastructure economically unreasonable. China's ability to offer superior price-per-performance, combined with its existing infrastructure and human resources cost advantages, makes it nearly impossible for the US to compete, even with faster chips. Professor Wong suggests the overall AI market is largely a 'huge bubble,' with 90% of its value being stock speculation, implying that its true economic impact might be overhyped. Nevertheless, this breakthrough is hailed as a win for science and humanity, making AI significantly more affordable and accessible globally, while simultaneously posing a significant geopolitical challenge to US economic and capital market stability.

"only artificial intelligence is capable of actually turning into a beneficial industry within a reasonable time frame"

"this llm AI is already considered I would say fourth or fifth generation of AI"

"the current generation of AI is a result of probability elimination when a user asks a question it is using a large uh quantity of online information data to generate a most likly answer it is not a result of active reasoning"

"to double the accuracy of an existing model you need to roughly push the computering volume by 1 million times"

"what the Deep seek team did is to bypass this physical mathematical constant and start to divide the AI into different modulated parts and train them separately"

"the cost of development and the cost of operation is only 1% of the other ones in the market right now"

"China didn't just catch up to the United States if China has free access uh to all those high performance chip deep seek will certainly surpass existing American companies"

"unless someone introduce a new generation of AI That's superior to the current generation LM the race for AI Supremacy within this generation is basically over the winner is China"

"Professor Wong even suggest that 90% of the value is just stock bubbles"

"this is definitely a win for Science and for us Collective human beings because AI will become a lot more affordable from now on"