Gianfranco's Best of July 2024 Reading List

Welcome to the July 2024 editing of my monthly reading list.

This month, I've curated my favorite essays across various categories: Artificial Intelligence, Society and Technology, Finance and Economics, Early Stage Investing, Writing, Life and Meaning, and Web3.

If you only have a few minutes, these three posts were my favorite, and are included in the list below:

The State of Chinese AI (Air Street Capital)

• Chinese AI labs are competitive, particularly in efficiency, but face a critical obstacle: advanced chip manufacturing. The essay highlights the near-impossibility of replicating ASML's cutting-edge lithography machines - as large as double-decker buses with over 100,000 precisely calibrated components. This hardware bottleneck contrasts sharply with Chinese firms' resourcefulness in acquiring AI chips, from creative interpretations of sanctions to luggage smuggling.

1.0 Is the Loneliest Number (Matt Mullenweg)

• In this 2010 post, WordPress cofounder Matt Mullenweg challenges the perfectionist approach, praising the release of imperfect version 1.0s. He argues that shipping early is "oxygen" for builders, creating a positive feedback loop of supporters who root for those who "build in public." This philosophy contrasts with the "one more thing" trap, which paralyzes development when "good enough" would suffice.

How to Be (Reasonably) Hard on Yourself (Nat Eliason)

• Complementing Mullenweg's essay, Nat Eliason provides insight into balancing self-criticism and creative work. He proposes a mantra: "It's good. It can be better. Eventually it will be good enough. Next time will be even better." This approach acknowledges that while the quality of one's current work can improve, perfection is unattainable. Eliason argues for consistent practice and iteration, explaining that mastery comes only from repeated effort over time, rather than a single grand attempt.

Artificial Intelligence

• Goldilocks” Agents and the Power of Custom Cognitive Architectures - Sequoia Capital

  • Excerpt: ““Focus on what makes your beer taste better,” making an analogy between how breweries at the turn of the 20th century made their own electricity and tech companies ran their own infrastructure before AWS. In a world where agents fall over more often than not, where 12-13% on SWE-bench is considered state-of-the-art, implementing a custom cognitive architecture absolutely makes your beer taste better.”

  • Read More

  • Pair with my essay on the same topic: Siri Gets An MBA (Published June 2023)

• The Four Barriers to AI Adoption - Tom Tunguz

  • Excerpt: “AI is new. Should a company allow a vendor to train a model using their data? Whose intellectual property is a fine-tuned model? What happens if a vendor violates the data privacy law? What training data is used that might subject the software buyer to future legal action?”

  • Read More

• Llama 3.1 405B, Meta’s AI strategy, and the new, open frontier model ecosystem - Interconnects.ai

  • Excerpt: “Meta’s strategy fully realizes that AI is a tool that they use and not one that is central to their business model, so they should commoditize their compliments. There is a long history of technology companies doing this — they’re weakening their competitors’ without kneecapping their own performance. When AI is moving so fast, it also has the potential of gaining the upside in tricks that the community finds for their models.”

  • Read More

Finance and Economics

• Working Title (Insurance) - Bits About Money

  • Excerpt: “To understand that, three magic insurance words you should know: frequency, severity, loss ratio. Frequency is the rate of occurrence of claims. Severity is the cost of claims contingent on claims happening. Loss ratio is the total amount of paid-out claims divided by collected premiums. Title insurance has extraordinarily low frequency for insurance products. However, when it does pay, the severity can be very high. Title insurance defenders will tell you that the reason title insurance is expensive is because the insurance company is promising to literally buy you a house in event of a problem.”

  • Read More

Fintech:

• How to Make Money Selling Money - Matt Brown @ Matrix Partners

  • Excerpt: “I’ve found the deeper and more powerful a risk insight is, the more confusing its manifestations in product and marketing can initially seem. This can be an “ugly” product. Seemingly weird ICP or product messaging. Features that should be there but aren’t, or that seemingly shouldn’t be but are. That’s because the company isn’t oriented around a traditional and observable approach to company building, but around an insight into risk that manifests in non-obvious ways.”

  • Read More

Early Stage Startups / Founder Guidance

• 1.0 Is the Loneliest Number - Matt Mullenweg

  • Excerpt: “By shipping early and often you have the unique competitive advantage of hearing from real people what they think of your work, which in best case helps you anticipate market direction, and in worst case gives you a few people rooting for you that you can email when your team pivots to a new idea. Nothing can recreate the crucible of real usage.”

  • Read More

•  Counterintuitive Advice for Building AI Products - Lenny’s Newsletter

  • Excerpt: “It’s actually easier and safer for startups to work on hard problems, problems that cannot quite be solved with today’s foundation models. We’re excited about riding the capability curve of improving models, instead of fighting that progress.”

  • Read More

• The Amazon Weekly Business Review - Commoncog

  • Excerpt: “Most businesses ask “how did the business do last week?” instead of “what did the customer experience last week?” If you ask the first question, you will have an implicit orientation towards internal, business-first metrics. Perhaps you might watch inventory turns, or track changes in free cash flow, or measure the performance of your warehouses. To be clear, these are all important metrics to track. But notice that if you ask “what did the customer experience last week?” you’ll measure slightly different things. And so it’s important to ask both questions, and to ask the ‘customer’ question first — it’s not for nothing that Amazon aims to be ‘Earth’s most customer-centric company.’”

  • Read More

• How to Interview and Hire ML/AI Engineers - Eugene Yan

  • Excerpt: “Sometimes, when software engineers first start working with ML models, they expect a level of control and predictability similar to databases or conventional APIs. But to their surprise, ML models are completely different beasts. They don’t have perfect accuracy, their predictions may change when the model is retrained on new data, they lack clear interpretability on how they arrive at outputs, and for large language models (LLMs), the output is stochastic where the same input can lead to different outputs.”

  • Read More

• Manage the What, Not the How - Molly Graham

  • Excerpt: “Neither of the cofounders managed the hours that people were working or where they worked; they didn’t manage how the work got done. Their management was all about alignment (do people understand what’s important, what we’re building, and why) and clear expectations (“we need to ship this feature by this date”).”

  • Read More

Achievements and Life

• How to Be (Reasonably) Hard on Yourself - Nat Eliason

  • Excerpt: “This is why it’s generally a waste to go after big goals like “run a marathon” or “publish a book” just to check them off your bucket list. You’ll exert a huge amount of effort only to produce the worst version of something that you’re capable of. It might still be good! But you only get really close to your ideal of good by getting more reps in.”

  • Read More

Science and Technology

• Will We Ever Get Fusion Power? - Construction Physics

  • Excerpt: “In the very early stages of a technology, when it barely works at all, a Moore’s Law rate of improvement — doubling in performance every two years — is in fact extremely poor. If a machine that barely works gets twice as good, it still only barely works. A jet engine that fails catastrophically after a few seconds of operation (as the early jet engines did) is still completely useless if you double the time it can run before it fails. Technology with near-zero performance must often get thousands or millions of times better, and many, perhaps most, technologies exhibit these rates of improvement early in their development.”

  • Read More

• Childhoods of Exceptional People - Henrik Karlsson

  • Excerpt: “James would model patterns of reasoning by thinking aloud and ask John Stuart to recreate his thought, imitating the thought patterns. He would give him increasingly complex tasks (books or ideas that he wanted John Stuart to summarize and articulate), then he would scaffold John Stuart by asking questions that helped him solve the task, and he would coach and give feedback on how to improve.”

  • Read More

• Skilled Immigration is a National Security Priority - Noahpinion

  • Excerpt: “Of course the U.S. needs to train its own skilled workers, and it should constantly be striving to improve its education system and to direct students toward the fields where they’re needed most. But at the end of the day the U.S. represents only 4.2% of the world’s population, while China represents 17.4%. China has a much bigger talent pool than America because it’s simply a much bigger country. If the U.S. wants to match China’s gigantic pool of human resources, it must supplement domestic talent by recruiting from abroad. Mathematically there’s simply no other option.”

  • Read More

• Three Breakthroughs Driving AI Forward, from Regie.ai Founder Srinath Sridhar - Ashu Garg

  • Excerpt: “Pakistani Prime Minister, Imran Khan, delivered wildly popular speeches in a recent election using AI-generated voice and video, all while he was serving jail time. “He basically won the popular vote,” Sri relates. “I think the history books will write it as the first time someone actually ran an entire election campaign with synthetic voice and synthetic videos.”

  • Read More

• The State of Chinese AI - Airstreet

  • Excerpt: “The sole producer of advanced machines, Netherlands-based ASML, is no longer selling its top-end equipment to Chinese companies. These are difficult to smuggle as they’re produced in low volumes, are very expensive, and can be the size of double-decker buses. Chinese companies have allegedly attempted industrial espionage, but these machines contain over 100,000 components, which have to be calibrated perfectly. This likely makes them the hardest single piece of equipment in the world to replicate.”

  • Read More

Interesting Charts: