Transcript

We live in a world where information is fragmented, yet we also have access to more information than ever before.

Q: What are the key mental models you rely on to understand how the world works? A: I'm a big believer in systems thinking. I read a book called Thinking in Systems. I'm also on the board of the Santa Fe Institute, which studies complexity theory. I’d describe complex systems as multivariable, nonlinear systems—very hard to predict. They can behave one way for a long time, then one variable changes and the whole system shifts. Weather, stock markets—these are examples. Consequences can be first, second, or third order. You can’t think linearly or focus on a single variable. A change here affects something there, which affects something else. You have to consider the whole system.

This helps avoid trouble because you can anticipate consequences you might otherwise discover too late. For example, someone at a large dating site thought longer profiles would increase engagement. They tested it—it worked—so they rolled it out. Months later they discovered it hurt conversions. That’s a second‑order effect. You must stay conscious of consequences and avoid becoming deterministic about a single metric.

Investing Journey and Key Influences

I started on Wall Street, not in venture capital, so I absorbed the usual influences: Peter Lynch (One Up on Wall Street), A Random Walk Down Wall Street, Buffett’s letters, Ben Graham, Howard Marks. These people spent their careers assembling and publishing their thoughts. I read everything.

That foundation in financial bedrock was incredibly valuable. Later, I met Bill Miller, who ran Legg Mason and beat the S&P for 15 years. He called himself a value investor yet was the largest shareholder of Amazon. His view: value means an asset is underpriced relative to what you think it will be worth in the future. If you believe in network effects, Amazon could grow at an “unreasonable” rate for a long time.

I’ve always thought Wall Street is the buyer of the product venture capitalists create. Liquidity eventually comes through M&A or IPO, and the price is set by that group. Even when you’re backing two people and a PowerPoint, you’re thinking about what the company will look like when it’s grown up and whether Wall Street will value it.

Knowing the Bedrock of an Industry

People skim. They want summaries. But deep knowledge differentiates you.

A partner of mine once bought a charity dinner with John Lasseter of Pixar. Lasseter served a 10‑course meal in his studio, each course tied to a classic cartoon he considered foundational to animation. He’d show the cartoon, explain it, and you’d realize how deeply he knew the history.

Another example: during a world chess tournament, they held a trivia contest during a break. Magnus Carlsen won—because he knew the history of chess.

We live in a world of fragmented information, but we also have unprecedented access. With LLMs, you can sit in your car for an hour and learn anything. More people should study the history of their field.

Picasso was a wildly successful realist painter by age 14. You wouldn’t guess that from his cubist work. Mastery of fundamentals matters.

Imagine interviewing at P&G or Pepsi. If you’re the one person who knows the masters of marketing, that’s wildly differentiating. It signals passion. If learning that history feels tedious, you’re probably in the wrong field.

Obsessive Learning in Founders

History isn’t always a common trait, but obsessive learning is. Disruptions that enable new technology waves happen on the edge. Entrepreneurs exploiting those edges—today AI, previously mobile—go home at night and read everything. They need to be in the top 1% of understanding.

When mobile apps first emerged, no engineers had written mobile apps. A few people dove in and figured it out. That required obsessive learning.

The Silent Edge

Young people naturally live on the edge. Incumbents struggle because embracing the edge may require admitting past decisions were wrong. But everyone should stay curious about the bleeding edge.

As a VC, I’m always afraid a new app will appear that I haven’t seen. I try everything. I have five premium AI accounts because I don’t want to miss anything.

Understanding the old stuff (history) and the new stuff (the edge) makes you a power player. If you understand the legends of marketing and TikTok, you stand out.

Surprising Ways I Use AI

People underestimate how much AI can do. You might ask it to identify the top 10 of something, then plan to study them yourself. But you can ask it to list pros and cons, rank them by one dimension, then rank them again by another. You can build later steps into the prompt.

Early on, I’d ask for numbers and then add them myself. Eventually I realized I could just ask it to do that too.

I like ChatGPT for its project structure and memory. For restaurants, I use Gemini because it has Google review data—you can ask what dishes people rave about or warn against. Coders swear by Claude. Some finance people prefer Perplexity. For deep research on unfamiliar companies or countries, some say Claude is better. It’s still a mix.

The Future of AI Models

Whether we end up with one dominant model or many niche models depends on how things play out. In coding, people already swap models. Cursor even lets users pick the model. As optimization and price become more important, swapping may increase.

But heavy regulation could push the industry toward oligopoly. Some players know this and are begging for regulation because it raises barriers—especially against Chinese open‑source models.

Global AI Regulation

If one country regulates AI heavily and another doesn’t, the unregulated one may accelerate faster. Copyright rules could create divergence. Chinese open‑source models don’t face the same constraints.

China has around 10 open‑source models. Their competitive dynamic is more intense because everyone chose open source. That system can innovate faster. Models learn from each other. You can have a model train or test another model.

Imagine two agricultural societies: • In one, farmers sell goods and go home. • In the other, farmers must share best practices. The second evolves faster.

Chinese models often open‑source weights and publish techniques. Western startups quietly fork these models. Many Silicon Valley companies use them.

AI’s Impact on Investing

If AI changes everything, how does it change investing? That’s an open debate.

If you believe models become near‑sentient, maybe one model does everything and vertical models disappear. I lean the other way. Workflows and data moats matter. Legal AI startups ingest case law deeply and build proprietary databases. Hard to replace that with a general model.

But big players are talking about going after verticals. It’s TBD.

Limitations of Current Training Approaches

We might be running out of data—“painting in the corners.” One powerful improvement method is hiring experts at thousands per hour to fine‑tune models. But human knowledge has limits.

Do we hit an asymptote? Or do models become superintelligent and surpass us?

Some point to AlphaGo’s famous unexpected move as proof models can innovate. Others argue that’s only possible in constrained environments where computers can search vast possibility spaces. Real‑world systems are too complex.

Self‑Driving Cars

Tesla FSD is “scary good.” I’d be comfortable sitting in the back seat now. But corner cases are hard to fathom. In a world where all cars were autonomous, it’d be easier. But humans jump in front of cars for fun because they know the cars will stop. That’s not good.

Non‑Consensus Opinions

Having spent 20 years in China, it’s hard for me to adopt the vilification mindset common in Washington and Silicon Valley. The U.S. is 3–5% of the global population. “American exceptionalism” sounds different to the other 95%.

Are We Overfunding the AI Buildout?

Hard to know. Five years ago, I wouldn’t have believed the “Magnificent Seven” would reach $3T valuations and then drop free cash flow from $50–100B to near zero to spend on capex.

Venture capitalists increasingly believe in increasing returns and power laws. That makes them more willing to invest aggressively. Losses before profitability have grown from billions (Amazon) to tens of billions (Uber) to far more now.

Circular deals—cloud providers funding AI companies who then spend the money back on cloud—inflate growth and delay corrections.

Burn rates today are enormous. It’s hard to know unit economics when spending that aggressively.

Retail Investors and Tokenization

There’s no shortage of funding today. Tokenization could change pricing, but it risks speculation and manipulation if done without disclosure rules.

If someone tokenized Stripe, the fluctuating token price could affect employees. One reason companies stay private is to avoid that chaos.

Robinhood tried something similar; companies argued it would be illegal.

Tokenizing real estate or IPO shares could be disruptive. The IPO process today is unfair—bankers pick the price and shareholders. A freshman CS and finance student would design a better system: match supply and demand anonymously, like an auction. Direct listings were a step in that direction, but Wall Street resisted.

Stablecoins could disrupt credit cards. Other countries have instant bank‑to‑bank transfers; the U.S. doesn’t, due to regulatory capture. Stablecoins bypass that.

Visa and Mastercard

Visa and Mastercard have some of the highest operating margins ever—around 60%. They’re duopolies created by banks. There’s no reason payments should cost 2–3%. China leapfrogged the U.S. with digital wallets (WeChat Pay, Alipay) because the government enabled instant transfers.

AI and Moody’s

Moody’s power is its status as a trusted standard. Even if AI does the analysis, Moody’s watermark matters. But AI could disrupt proxy advisory firms like ISS, which score companies using opaque methods and get paid on both sides.

Passive Investing and Second‑Order Effects

Index funds own huge percentages of companies. If they didn’t vote, active shareholders would have more influence. Some argue index funds should vote proportionally to direct holders.

Passive dominance pushed active managers into “closet indexing.” But with fewer active investors, some argue the opportunity for alpha has increased. Hard to know.

The Craft of Storytelling

Storytelling is one of the top three traits of successful founders. I fell in love with long‑form nonfiction journalism—Gladwell, Michael Lewis, Krakauer. Their writing reads like fiction. Books like The New Journalism analyze that craft.

Writing forces clarity. Bezos uses six‑page memos for this reason. In venture, writing also attracts founders—it becomes a calling card.

Founder Advantages

Key advantages include:

• Product instinct — extremely hard to teach.

• Storytelling — founders are always selling: employees, investors, customers, partners.

• Determination — Bezos once said he asks only one question: Is this person going to do this no matter what?

Lessons from Uber

Uber operated in winner‑take‑all dynamics with massive network effects. Burn rates escalated to levels no public company would tolerate. There was no HBS case study for that situation. No mentors. Now AI companies face similar dynamics.

Inside Benchmark’s Structure

Benchmark is an equal partnership—no hierarchy, no CEO. Consequences:

Positives:

• Easy to recruit exceptional talent.

• Strong development of new partners.

• No internal politics around compensation.

• Partners freely share resources because everyone benefits equally.

Negative:

• Hard to scale or launch new initiatives because no one “owns” them.

• Even the website became a challenge—eventually replaced with a simple splash page.

Why Founders Choose Benchmark

Success attracts founders. Reputation creates network effects. Founders want partners who understand and are excited about what they’re building. Young VCs can outperform because they understand new technologies and niches deeply.

Venture skews young because it’s a hustle business—constant learning, constant exploration.

What Is Success?

Success has changed for me. I left venture because I felt there was no work left to do. It was my dream job; I loved it. Now I want to apply the techniques I used in venture—writing, synthesizing, understanding problems—to broader societal issues and see if I can “dent the universe” in a new way.