Why Correct Predictions Are Not Enough: Survival Comes First

Recently, I came across a news headline that caught my attention: “Why Did the 25-Year-Old AI Genius of Wall Street, Leopold Aschenbrenner, Fail?” The headline prompted me to write this blog post because I have seen friends go through somewhat similar experiences in their own investing. I should also mention that I am not a professional investor. What follows is simply my personal attempt to think about investing through the lens of mathematics, with substantial assistance from ChatGPT.

On July 30, 2026, Leopold Aschenbrenner, the 25-year-old founder of the San Francisco-based hedge fund *Situational Awareness*, sent a letter to his investors. A former OpenAI researcher, Aschenbrenner graduated from Columbia University in 2021 as valedictorian with a B.A. in economics, mathematics, and statistics. In the letter, he acknowledged that his fund had lost **67% of its value in July alone**. For someone who had been widely regarded as an exceptionally talented young thinker in AI and investing, the disclosure was striking and sobering.

Aschenbrenner is clearly intellectually gifted and recognized one of the most important technological trends of our time relatively early. Yet exceptional academic ability and broad insight into AI do not necessarily translate into a deep understanding of every aspect of deep learning, nor do they automatically translate into sound investment judgment. His widely discussed *Situational Awareness* report contains many important observations, but much of its discussion of AI development is relatively high-level, and it is not always clear how deeply its projections incorporate the technical limitations and uncertainties of current deep-learning approaches. More importantly, being right about a technological trend is very different from successfully investing in it.

This distinction becomes clearer when Aschenbrenner is compared with Warren Buffett. Aschenbrenner may represent good analytical ability combined with relatively limited practical investment experience. Buffett, by contrast, has accumulated more than six decades of experience navigating market cycles, mistakes, uncertainty, and changing economic conditions. Successful investing requires not only identifying attractive opportunities, but also managing risk and remaining invested long enough for compounding to work. From a mathematical perspective, Aschenbrenner’s approach can be viewed as placing greater emphasis on expected return, while Buffett’s philosophy can be interpreted as placing greater emphasis on long-run compounded wealth. These objectives may appear similar, but they can lead to meaningfully different investment decisions.

Suppose an investor's wealth evolves according to $W_{t+1}=W_t(1+r_t),$ where $r_t$ denotes the return during period $t$. Most investors naturally try to maximize the expected return, $E[r].$ This way of thinking is perfectly intuitive. If one correctly identifies the most transformative technology of the century, the rational response seems to be maximizing exposure to that opportunity. This was essentially Aschenbrenner's way. His investment thesis was intellectually compelling. Artificial intelligence would require enormous computational resources, HBM supply would remain structurally constrained, data centers would become strategic assets, and companies supplying AI infrastructure would therefore generate exceptional returns.

The industrial thesis may ultimately prove correct. The mathematics of leverage, however, tells a very different story. Suppose an investor employs three times leverage. A decline of only 20% in the underlying assets immediately translates into an approximately 60% loss of equity. If losses continue, the investor may face margin calls and forced liquidation. Mathematically, $W\rightarrow0,$ and once wealth approaches zero, every future investment opportunity becomes irrelevant. Regardless of how accurate the original thesis was, the investment process has effectively ended.

Buffett approaches the same problem from an entirely different perspective. Rather than focusing solely on maximizing $E[r],$his investment philosophy can be interpreted as being much closer to maximizing $E[\log W],$ which is the objective underlying the Kelly Criterion. Buffett has never claimed to follow the Kelly Criterion explicitly, yet many of his investment principles are remarkably consistent with its mathematical intuition.

The logarithmic utility function has an important implication. Wealth compounds multiplicatively rather than additively, making gains and losses fundamentally asymmetric. A gain of 100% doubles wealth, whereas a loss of 50% cuts wealth in half. Recovering from that loss requires another gain of 100%, not 50%. Likewise, an 80% loss requires a subsequent gain of 400% merely to return to the original capital. This asymmetry explains why avoiding catastrophic losses is mathematically far more important than achieving spectacular gains.  

The same idea becomes even more transparent when viewed in terms of logarithmic returns. A 50% loss corresponds to a wealth multiplier of $0.5$, giving a logarithmic return of $\log(0.5)=-\log 2.$ To recover to the original level of wealth, the portfolio must subsequently double, corresponding to a wealth multiplier of $2$ and a logarithmic return of $\log 2.$ Hence, $\log(0.5)+\log(2)=0.$ In other words, losses and recoveries are perfectly symmetric in logarithmic space, even though they are highly asymmetric when expressed as percentage returns. This is one reason why logarithmic wealth plays such a fundamental role in the mathematics of long-term investing. It naturally reflects the multiplicative nature of compounding and highlights the importance of avoiding large drawdowns.

The distinction becomes even clearer when viewed through the lens of long-term compounding. Wealth after $n$ investment periods is given by $W_n=W_0\prod_{i=1}^{n}(1+r_i).$ Investment success is therefore governed by a product rather than a sum. One sufficiently large negative event can permanently impair decades of successful investing. Long-term wealth depends not only on earning high returns, but also on remaining in the game long enough for compounding to work.

Viewed from this perspective, Aschenbrenner optimized the magnitude of future gains, whereas Buffett optimized the probability of surviving every market cycle. Conceptually, their objective functions can be written as Aschenbrenner:$\max E[r]$ and Buffett: $\max E[\log W].$ Perhaps the most revealing way to summarize Buffett's philosophy is through the following conceptual equation:  Investment Success=Correct Thesis $\times$ Position Size $\times$ Survival Probability.

Many investors devote most of their attention to the first term. Buffett appears to devote substantial attention to the last two as well. He is often willing to give up part of the potential upside if doing so increases the likelihood of remaining financially secure through different market conditions. Although this approach may appear conservative over shorter periods, it has contributed to an exceptional long-term investment record. Aschenbrenner may ultimately be remembered as an insightful analyst of the AI revolution. His views on semiconductor supply chains, HBM, and AI infrastructure may prove broadly correct. Yet financial markets reward not only good predictions, but also portfolios that can withstand uncertainty, volatility, and imperfect timing.

This may be the most important mathematical distinction between the two approaches. Aschenbrenner attempted to identify the future early. Buffett built an investment system designed to remain resilient across many possible futures.

The lesson extends beyond these two individuals. In investing, making a correct prediction is valuable, but it may not be sufficient. Some investors are eventually proven right but still fail because they run out of capital, liquidity, or time before their thesis is realized. Intelligence can help identify opportunities, but financial resilience determines whether those opportunities have time to compound.  In the end, mathematics suggests something simple: the first condition for compounding is to remain in the game.

Perhaps the most valuable education is not learning how to be right, but learning how to survive being wrong. Small, survivable mistakes cultivate humility and judgment. The objective is not to avoid every mistake, but to ensure that no single mistake is large enough to end the game.

This distinction also highlights an important misunderstanding of the phrase "high risk, high return." In academic research, a high-risk project usually means that the research itself may fail, while the researcher remains able to continue working and pursue the next idea. In business, entrepreneurship, or investing, however, high risk can mean something fundamentally different: a single mistake may destroy the organization or permanently eliminate the opportunity to continue. In those settings, survival is itself a strategic objective.