Hey Reader - Thomas here.

Good day to my fellow compounders!

In 3-Bullet Sunday, I share with you the three most interesting pieces of articles, videos, or podcasts I came across for the week on investing and growth philosophies.

Enjoy!

Why We Choose Steady Compounding This week, a hedge fund that was up 439% for the year nearly blew up. Situational Awareness is an AI-focused fund run by a former OpenAI researcher. It launched less than two years ago, and by the start of July it had swelled to $45 billion in assets. Then AI stocks sold off hard, and because the fund was reportedly running around four times leverage, one bad month was all it took. The margin calls came in, and the entire public stock portfolio was sold to Ken Griffin's Citadel in a single block. Even after all that, the fund is still up roughly 80% this year (likely due to their private equity position, Anthropic, that's offsetting the catastrophic cash losses from the public market bloodbath... but this is just my guess). It just very nearly didn't survive to enjoy it. Reading the news, my first reaction was that this story feels familiar. We have seen it before. In March 2021 it was Archegos, Bill Hwang's family office. Concentrated bets on tech and media stocks, roughly five times leverage through swaps, and when the positions fell, $20 billion of wealth evaporated in two days. (Hwang was later convicted of fraud over how those positions were built, which is its own saga. The overlap with this week is the leverage and the margin calls, not the fraud.) In 2022 it was Three Arrows Capital, the crypto fund run out of Singapore. Same shape. Riding the sector bull wave, leveraged up, gone within weeks once prices turned. And in 1998 it was Long-Term Capital Management, which had two Nobel Prize winners on the team and still needed the Federal Reserve to organise a rescue because of, you guessed it, leverage. And it isn't just the famous cases. I know a few people personally who went through their own version of this. A bull market came along, they got intoxicated with the easy money, started taking on leverage to speed things up, and had their ass handed back to them when a speedbump came along. I understand the temptation. When your portfolio grinds out a normal year while some AI fund is printing 400%, the slow path feels almost embarrassing, and leverage looks like the fastest way to catch up. In a bull market it even looks free, because every month it works, it feels a little smarter. These funds all invested in different assets. What blew them up was the same every time: leverage. If the first half of this year made your portfolio feel slow, I hope this week serves as a reminder of why we invest the way we do. No borrowed money means nobody can ever force us to sell. It isn't the fastest way to compound, but it keeps us in the game long enough for compounding to work. That is also why I wrote The Lunch Break Investor. It's the slow path in book form, a six-step system for analysing and owning great businesses one lunch break at a time. The book launches on August 18, and you can pre-order it here. Compound Steadily, Thomas P.S. If you're in Singapore, I'm doing a book talk at Kinokuniya Takashimaya (Level 4) on 22 August, 4pm, with He Ruiming from The Woke Salaryman moderating. Come say hi: details here.

​

I. Warren Buffett Investing Process

Alice Schroeder is Warren Buffett's biographer, and she shared about his decision-making process in 2008 at the Value Investing Conference.

It provided an invaluable peek behind Buffett’s decision-making process when evaluating a new business.

​>> Click here to read the transcript​

Some of my highlights:

If there's a good probability of catastrophe risk, it's an immediate no

Warren doesn't use a model at all, no projected earnings or returns.

Here's what he did instead—identify the handful of factors that could make or break the thesis.

Buffett's "DCF" model—15% returns from day one

II. Satya Nadella | BG2 w/ Bill Gurley & Brad Gerstner

​

Some good insights into AI and how its reshaping the tech world.

Some of my highlights:

AI Arms Race

• Competitive Landscape: Nadella acknowledged the competitive landscape of the AI industry, with major players like Google, Amazon, Meta, and now Elon Musk's xAI entering the fray. He emphasized that, unlike previous tech shifts, all the major players are now actively participating in the AI revolution
• Not Winner-Take-All: Nadella believes that the AI industry won't be winner-take-all. He anticipates multiple winners at different layers of the AI stack, including infrastructure, models, and applications.

Legacy Search (Google) and Consumer AI

• Changing User Habits: Nadella discussed how consumer habits are changing with the rise of AI. He noted that younger generations prefer direct answers from AI agents like ChatGPT over traditional search engines.
• Bing's Role: Nadella emphasized the importance of Bing and the broader search business, even in the age of AI. He highlighted Microsoft's efforts to integrate AI into Bing and develop new AI-powered consumer products.
• Distribution Challenges: Nadella recognized Google's distribution advantages in the search market, particularly on mobile devices. He stressed the need for Microsoft to innovate and leverage its strengths, such as the Windows operating system, to compete effectively.

The Future of AI Agents

• Agentic Interactions: Nadella discussed the rise of "agentic" interactions in enterprise settings, where AI agents can seamlessly interact with various business applications and data sources. He envisioned a future where AI agents would handle scheduling, communication, and data analysis, significantly improving productivity.
• Permission and Control: Nadella emphasized the importance of user permission and control in the context of AI agents accessing and interacting with applications and data. He highlighted the need for clear guidelines and security protocols to ensure responsible AI implementation.

Leveraging AI within Microsoft

• AI for Productivity: Nadella discussed how Microsoft is leveraging AI internally to improve productivity and efficiency. He cited examples like using AI for customer service, code development with GitHub Copilot, and streamlining workflows across various departments.

The Cost of Model Scaling and Inference

• Scaling Laws and Economic Realities: Nadella discussed the balance between the benefits of scaling AI models and the economic realities of training and inference costs. He noted that while scaling laws suggest performance improvements with larger models, practical limits and economic considerations exist.
• Optimizing Inference: Nadella highlighted the importance of optimizing AI inference to improve efficiency and reduce costs. He discussed techniques like test-time computing and diverse hardware architectures to balance performance and cost.

III. ($) Lululemon: Market is Optimistic About Q4 and Beyond

This week's Insider Stock article is on Lululemon.

Members, login here and read my updated thoughts on Lululemon Q3 results

Not a member yet?

​>> Click here to join and access my entire archive of stock research​

That's all I have for you today!

Till next week.

Thomas Founder of Steady Compounding Connect with me: ​​​​

​