“When the facts change, I change my opinion. What do you do, sir?” This adage, often attributed to economist John Maynard Keynes, underscores the virtue of adaptability in the face of new evidence and an epitome of intellectual honesty, if not submission to the vagaries of real life. Yet, in the unfolding saga of DeepSeek, arguably the most significant development in the AI landscape in years, we observe a contrasting phenomenon: a steadfast clinging to preconceived notions, with justifications conveniently reshaped to fit emerging facts. 

From our pioneering POV

The sudden explosion of interest in DeepSeek forced commentators from every walk of life—many with little prior engagement in AI—to scramble for an understanding of something entirely alien to them. We, on the other hand, had the luxury of stepping back and analyzing the unfolding reactions, having already done the work weeks before:

• We were likely the first in the global financial industry to recognize DeepSeek's potential, writing about it as the most important announcement ahead of all the announcements from OpenAI or Google’s Quantum announcement, shortly after its Christmas release. Having published the paper on 2 Jan at  https://geninnov.ai/s/XBzlRD, we deem it unlikely anyone in the finance industry globally did it before us!
• Frustrated by the lack of attention DeepSeek was receiving in the financial world, we published a detailed analysis of its impressive features on January 9th, 2025, well before it caught mainstream attention even in name. We believe we were the first in this, too, but also perhaps the most comprehensive in listing the features, partly because of our background and partly because we were not under time pressure! At https://geninnov.ai/s/fSKrWw
• Prompted by a savvy reader for concerns about its impact on NVIDIA and AI hardware stocks, we conducted a rigorous analysis of AI compute demand. Our calculations showed that inference, not pre-training, drives hardware needs, reinforcing why DeepSeek’s efficiency wouldn’t fundamentally weaken demand for AI chips. Again, we had to be the first as there was little concern about DeepSeek in markets even at this point and almost no mention amongst finance world commentators - our article at https://geninnov.ai/s/YAaXZr

We are upset we did not use the term “Jevan’s Paradox.” That’s a fancy term.

The Shifting Sands of Argument

As DeepSeek's prominence grew, drawing attention from even the most tech-averse commentators, a fascinating pattern emerged. DeepSeek became a convenient tool to bolster existing narratives. Many are hyperventilating about a Sputnik moment, but we are yet to notice anyone changing recommendations, market stances, or their overall views on GenAI or innovation. There are numerous new arguments, but little opinion changes.

For instance, we have read multiple macro strategists who have been consistently advocating for a reversal in tech stocks’ outperformance and now see DeepSeek's achievements to change the justifications for the tech collapse. Similarly, analysts who previously downplayed the AI revolution on capabilities and being too expensive now highlight the problems of the technology emerging as too cheap, even if more capable and universal. Sector analysts, most notably from the software, who have been hurting with the AI themes, have come in with their DeepSeek angles to promote the positive stance on their segments.

After a day under the covers, the hardware and energy sector fans, began to find their footing. The first natural response was skepticism toward DeepSeek, often manifested as accusations of "deepfakes" or a lack of transparency. The model's open-source nature and undeniable capabilities forced a reassessment, with many now embracing the positive implications for increased AI usage and hardware demand. This is effectively our camp.

This is why we need to examine our behavioral blind spots. A big fact has emerged, and we, like everyone else, have not changed our opinions. This tendency to adjust justifications rather than conclusions underscores the importance of establishing clear benchmarks for belief revision. In the following sections, we outline our core beliefs about the AI landscape and the specific "goalposts" that would compel us to reassess our views.

AI's Exponential Growth

Core Belief: We hold that AI capabilities will continue to rise exponentially, driving transformative changes across industries.
Benchmark for Reevaluation: A noticeable decline in the pace of innovation, evidenced by a lack of significant new features or products from leading AI companies over an extended period, would prompt us to reassess this belief. We are unlikely to be swayed by hypothetical arguments like Scaling Laws or anecdotal tales of the doubters, who will always turn up with modified examples of where the models do not work. 

Innovation as a Global Endeavor

Core Belief: Innovation is a global race, with significant contributions emerging from regions beyond the U.S., particularly in Asia.
Benchmark for Reevaluation: A prolonged absence of notable innovations or technological advancements from Asian countries would lead us to reconsider the global distribution of innovation. For now, DeepSeek has bolstered the belief.

Application Layer Dynamics

Core Belief: The software or the application layer is under pressure due to rapid feature replication and diminishing unique value propositions.
Benchmark for Reevaluation: If application-layer companies consistently develop unique products and features that resist replication and command premium pricing, we would revisit our stance on their market position. We need evidence not through the announcements of new GenAI-driven products, services or features but examples of features that do not immediately face replication and improvement in our instant copiability world. For now, the blurring of application boundaries has created a highly competitive environment where unique and sustainable advantages are difficult to achieve. DeepSeek has only shown the immediate competitive pressure on the highly-priced, premium features of OpenAI. 

Innovation as a Growth Imperative

Core Belief: Sustained growth in the coming years will favor innovators as traditional drivers become less effective. This will make innovation investment the best growth investment in any portfolio. Innovations are not just GenAI, but across sectors due to AI-created changes and also through newly emerging industries like Robotics.
Benchmark for Reevaluation: For us, the key risk is when AI model innovations fail to impress, and consequent innovations in biotech, mobility, material sciences, and Robotics take a few years to emerge. 

Conclusion: Embracing Adaptive Thinking

The above is an illustrative list. Given the observations on justification changes, we are committed to establishing clear benchmarks for our assumptions and conclusions, both at macro and micro levels. This is for ourselves and our processes. We do not need to think about others, but for ourselves, we must be perpetually aware of the importance of remaining open to change and not getting stuck in the quagmire of our biases and announced conclusions. 

Equally importantly, The DeepSeek saga reminds us that in the world of innovation, nothing is guaranteed. Technological progress can be unpredictable and defy expectations. It is important to stay evidence-based rather than expect the innovation world to evolve along any theoretical law-based projections or to fit any historical patterns, however impressive or consistent.