2025 October Digest

Here's my 2025 October Digest.

The Art of Spending Money

This is Morgan Housel’s latest book, considered a sequel to his blockbuster The Psychology of Money. Like his first book, it explores the intersection of personal finance and psychology, specifically how our mindset influences our saving and spending habits. The book is filled with timeless wisdom, shared through engaging stories. It’s a great read, and I highly recommend it. Here are some highlights:

• The Most Valuable Financial Skill: "Not needing to impress other people." Eliminating this desire is described as an asset worth a billion dollars, as it frees up capital that would otherwise be spent on status signaling. "Nobody is thinking about you as much as you are. Nobody cares about your stuff as much as you do."
• True Wealth is Autonomy: Wealth is not defined by the ability to spend money, but by the unspent assets that grant independence and control over one's time. This autonomy is a primary driver of happiness and well-being. "Every bit of savings that you have is a piece of your future that you own." This reframes the purpose of saving from mere accumulation to the direct purchase of autonomy. The ultimate goal is not a Lamborghini but the ability to wake up every morning and have complete control over one's day. This is encapsulated in a quote from Charlie Munger: "I never wanted to become rich, I just wanted to become independent."
• Endurance is the Ultimate Investment Skill: The most powerful force in wealth creation is not achieving high annual returns, but the ability to earn average returns over an exceptionally long period. Patience and the discipline to stay invested through market volatility are far more important than genius stock-picking. This reminds me of Mohnish Pabrai’s breakdown of the three elements of wealth building: starting capital, length of the runway, and rate of return. Endurance as the ultimate investment skill reinforces the importance of patience and maintaining a long investment horizon.
• Happiness is the Gap Between Expectations and Reality: Financial satisfaction is determined less by income and more by managing one's expectations. The constant upward creep of expectations, fueled by social comparison, is a primary cause of unhappiness, regardless of wealth.
• Risk is What You Don't See: The greatest financial threats are the unpredictable, black-swan events that are never forecasted. The proper response is not prediction but preparedness—maintaining a significant buffer (e.g., cash) to survive any eventuality.
• Making Money and Keeping Money Are Different Skills: Gaining wealth often requires optimism and risk-taking. Preserving it requires humility, paranoia, and a focus on survival. The failure to distinguish between these two skill sets is a common path to ruin.

Andrej Karpathy's Interview

One of the best researchers and teachers of Deep Learning and AI of our time, Andrej Karpathy always commands attention in in-depth interviews. In this one, he explains why reinforcement learning is terrible (though, in his view, everything else is even worse), why AGI will likely blend into the past ~2.5 centuries of ~2% GDP growth, why self-driving took so long to crack, and what he envisions for the future of education.

• Our current way of building AI: He posits that we are "summoning ghosts, not building animals," creating ethereal, digital intelligences through imitation of internet data, a process fundamentally different from biological evolution. This leads to models with powerful but flawed capabilities. Current models possess significant "cognitive deficits"—lacking continual learning, robust reasoning, and a mechanism for distilling experience—and are overly reliant on memorization, which he views as a bug, not a feature.
• On Reinforcement Learning: Karpathy is highly critical of current Reinforcement Learning (RL) techniques, describing them as “terrible” and likening them to “sucking supervision through a straw” due to their noisy and inefficient reward mechanisms. The core problem with outcome-based RL is its extreme inefficiency and noise. An agent may try hundreds of different trajectories to solve a problem (e.g., a math problem). When a few trajectories succeed, every single action taken within them is upweighted, regardless of whether it was a brilliant step or a mistake that was later corrected.
• The Practical Realities of AI-Assisted Programming: He identifies that AI assistants still fall short on tasks that require a lot of design. This is because agents are biased by common patterns from their training data. It produces a lot of "slop" - where the generated code is often bloated, overly defensive (excessive try-catch blocks), and uses deprecated APIs. The fact that models are asymmetrically worse at writing "code that has never been written before" is a direct counterargument to scenarios of a rapid, recursive intelligence explosion.
• AI, Superintelligence, and Economic Impact: Karpathy views AI not as a separate, world-altering technology but as an extension of computing and automation—a trend that has been underway for centuries. He argues that we are already in an intelligence explosion and have been for decades, visible in the exponential curve of GDP. He pushes back against the idea that AGI will cause a "discrete jump" in the economic growth rate.
• Future of Work: He predicts that jobs will not be replaced instantly. Instead, an "autonomy slider" will gradually increase, with AIs handling more of the rote work (e.g., 80% of call center volume) while humans move to supervisory roles. The most likely long-term outcome of superintelligence is not a single rogue AI but a "gradual loss of control and understanding" as society becomes dependent on a complex "hot pot of completely autonomous activity" involving multiple competing AI entities acting on behalf of different human interests.
• A Vision for Education in the Age of AI: He frames education as a "very difficult technical process of building ramps to knowledge." The objective is to create learning materials that maximize "eurekas per second" by untangling complex topics and presenting them in a sequence where each concept builds logically on the last. His teaching approach is rooted in his physics background: finding the first-order approximation, building simple models ("spherical cow"), and isolating the core intellectual idea 

Note to My Slightly Older Self

A refreshing read from Yew Jin Lim on his lessons learned that he want to share with his 30+ year old self after surviving that phase. A few nice take aways:

• Lead with vision, not tasks: Managers keep things running. Leaders change where things are going. Paint the destination, not the path.
• Do the next job now: Act at the next level now, instead of asking for that next promotion. Have the career conversation with yourself quarterly, don't want for performance review. Document your scope, expansion and learnings. Make sure what you work on is on the critical paths of your manager and the organization.
• You are your own safety net: As a mid-career professional, you might have multiple successful projects to point to, skills marketable anywhere, a network built over years of being helpful without agenda, and most importantly, the experience of having failed, recovered, and lived to tell about it. The last one is the most important asset of yours. That's where resilience come from.
• One Miracle per project: When evaluating any initiative, map out everything: technical challenges, resources, timeline, dependencies, org changes. Here’s the rule: You get one “miracle” - one major unknown you’ll need to solve. That’s your innovation space. Too many projects spiral because they needed multiple breakthroughs to succeed. The most successful projects weren’t the most ambitious - they found that sweet spot between innovation and realistic execution.
• Define success yourself: Choose your own scorecard, you define it. Stop apologizing for the choices you make, own them and be proud of them. Re-evaluate your scorecards and choices regularly and course correct.