Gianfranco's Best of November 2024 Reading List

Welcome to the November 2024 edition of my monthly reading list.

This month, I've curated my favorite essays across various categories: Semiconductors, Artificial Intelligence, Business and Technology, Science and Society, Finance and Economics, and Personal Development.

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

• A Chance to Build - Ben Thompson
  • Thompson unpacks the unspoken history of Silicon Valley, revealing how the "zero marginal cost" myth of chips masked a decades-long reliance on Asian manufacturing, fueled by government policies and global economic structures. He connects this to the current geopolitical landscape, highlighting how Trump's tariff proposals aim to disrupt this delicate balance. The core question: can tariffs truly revitalize U.S. manufacturing, or will they cripple the software-hardware symbiosis that has propelled tech's growth? The stakes are high: disrupting this global dance could reshape everything from AI development to the future of self-driving cars, forcing tech companies to play the game on a very different field.
  • Read More
• Everything I've learned so far about running local LLMs - Chris Wellons
  • Wellons offers a refreshingly candid and technically grounded exploration of the burgeoning world of local LLMs. He tackles the hype head-on, sharing practical tips on model selection, quantization, and the nuances of fill-in-the-middle prompting. The key takeaway: while running LLMs locally bypasses vendor lock-in and opens up exciting possibilities, the tech remains nascent. LLMs might not be coding software engineer jobs out of existence just yet, but their potential for tasks like proofreading and creative writing is undeniable, reminding us that the real revolution may not be in replacing humans, but in augmenting our capabilities.
  • Read More
• Maximizing Time for Reading - Blake Butler
  • Butler argues for the enduring importance of reading widely, especially in our current age of digital distraction. He dismisses the "elitism" critique, emphasizing that reading is a practice accessible to anyone willing to invest the time. He provides practical tips for carving out reading time, from utilizing small pockets of downtime to integrating reading with other activities. The key takeaway: Reading is not just a pastime; it's a vital form of nourishment for the mind, body, and soul, offering a deeper engagement with the world and a connection to the larger human experience.
  • Read More

Semiconductors

• A Chance to Build - Stratechery
  • The article unpacks the global semiconductor industry's evolution, illustrating how U.S. innovation and Asian manufacturing expertise created one of the most potent economic systems in history. However, shifting geopolitics, tariffs, and automation are challenging this balance, opening opportunities for reshoring high-tech manufacturing powered by robotics and AI. The key takeaway: the U.S. must rethink manufacturing as a modular, software-defined, and automation-driven process to compete in a reshaped global economy.
  • Read More
• The Rise of Amazon's Trainium 2 ($) - Fabricated Knowledge
  • Amazon is doubling down on custom silicon with Trainium 2, signaling a strategic pivot toward GenAI and a serious attempt to challenge Nvidia’s dominance in AI infrastructure. This shift in AWS’s capital investment into proprietary AI chips, projected to outpace past growth rates, could mark a new era of competition, driving costs down and accelerating development cycles across the hyperscale sector.
  • Read More
• Profit Dollars per GPU Dollar - Tom Tunguz
  • AI’s acceleration is reshaping the cloud landscape, with AWS, Azure, and GCP reporting triple-digit AI growth but facing a capex crunch driven by surging GPU demand. As these giants build custom chips to cut costs, the real contest is emerging around ‘profit per GPU dollar’—whoever masters this metric will likely dominate the AI-powered cloud market.
  • Read More
• Does All Semiconductor Manufacturing Depend on Spruce Pine Quartz? - Construction Physics
  • While Spruce Pine, NC supplies most (70-90%) of the semiconductor industry's crucial high-purity quartz for crucibles, alternatives exist but face cost and technical limitations. The real bottleneck isn't the mine itself but rather the quartz crucible technology, whose limitations significantly impact manufacturing costs and efficiency, especially for solar PV production.
  • Read More

Artificial Intelligence

• OpenAI Shifts Strategy as Rate of 'GPT' AI Improvements Slows - The Information
  • OpenAI's upcoming Orion model shows diminishing returns in performance gains compared to previous generational leaps, forcing the company to pivot toward post-training improvements and reasoning models. This technological plateau suggests the industry may be hitting fundamental scaling limits with traditional training approaches.
  • Read More
• Everything I've Learned So Far About Running Local LLMs - nullprogram.com
  • Local large language models have become accessible enough to run on consumer hardware, with a 4GB model outperforming the original ChatGPT possible on a Raspberry Pi, though practical limitations around VRAM, model size, and reliability create clear boundaries for real-world use. Despite rapid progress in local deployment, LLMs remain best suited for specific tasks like proofreading and creative writing, while struggling with code generation and tasks requiring deep context or perfect accuracy.
  • Read More
• Small But Mighty AI - Tom Tunguz
  • Despite the hype around massive AI models, a recent Databricks survey reveals that 77% of enterprise AI deployments use smaller models (under 13 billion parameters). This preference stems not only from significantly lower costs and latency, but also from the impressive performance gains of smaller models, making them a compelling choice for businesses seeking practical AI solutions.
  • Read More

Business and Technology

• Meta Earnings, Measuring Meta’s Spending, Meta Recommendations - Stratechery
  • Meta’s AI and capex-driven growth strategy remains a calculated gamble on long-term payoff, as Zuckerberg pushes heavy investment into AI across the platform while doubling down on Reality Labs, despite increasing costs and investor skepticism. Zuckerberg’s pitch: Meta’s AI is foundational, influencing everything from ads to Ray-Ban smart glasses, with revenue as the ultimate test of success. For investors, it’s clear—either trust Meta’s expensive, far-reaching AI vision, or step off the train.
  • Read More
• Why Middlemen Don’t Get Eliminated - The Diff
  • While the internet promised disintermediation, powerful middlemen like Amazon and Google now dominate the digital economy. This is because cheaper information creates more niche markets and products, increasing the value of intermediaries who efficiently connect consumers with precisely what they want. As a result, the relative advantage of selling through these platforms outweighs the cost, even though going direct remains an option.
  • Read More
• Trump’s Banks ($) - Net Interest
  • Major financial institutions—from mortgage giants Fannie/Freddie to crypto firms and traditional banks—saw significant stock gains following Trump's election, driven by expectations of lighter regulation and privatization opportunities. The anticipated policy shifts could reshape the financial landscape through potential privatization of Fannie/Freddie, easier M&A approvals, crypto-friendly policies, and relaxed bank capital requirements.
  • Read More
• Oracle: Ellison's Voyage to Software and Beyond - MBI Deep Dives
  • Oracle's journey, driven by Larry Ellison's relentless competitive spirit, started with a revolutionary idea about relational databases but evolved through aggressive sales tactics, strategic acquisitions, and a bumpy transition to the cloud. The company's ability to navigate market shifts, from the initial database wars to the current AI revolution, demonstrates its enduring relevance.
  • Read More
• 2024: The State of Generative AI in the Enterprise - Menlo Ventures
  • Enterprise adoption of generative AI is accelerating, shifting from pilot programs to production deployments, with spending increasing sixfold. While focused on practical use cases like code generation and chatbots, enterprises are strategically diversifying across multiple foundation models and increasingly building in-house solutions. Incumbent vendors face growing dissatisfaction, opening doors for specialized, AI-native startups.
  • Read More
• Betting on and Against Feedback Loops - The Diff
  • MicroStrategy's unusual strategy of issuing equity to buy Bitcoin creates a reflexive feedback loop where stock price appreciation fuels further Bitcoin purchases, leading to a persistent premium valuation. While shorting this dynamic is challenging due to the self-reinforcing nature of the loop, the trade's reliance on sustained investor enthusiasm creates vulnerability to waning interest. This makes MicroStrategy a compelling, albeit complex, case study in valuation-driven short selling.
  • Read More

Science and Society

• The Present Future: AI's Impact Long Before Superintelligence - One Useful Thing
  • AI’s present capabilities—far from needing future superintelligence—are already reshaping work by automating tasks and introducing new forms of oversight. While AI can serve as an observant manager or helpful assistant, it risks becoming a tool for invasive monitoring.
  • Read More
• The Wisdom and Madness of Crowds: Market Prices as Political Predictors - Aswath Damodaran
  • Political prediction markets like Polymarket and Kalshi emerged as alternatives to traditional polling in the 2024 election, offering real-time, money-backed forecasts that proved more dynamic than polls but suffered from volatility and potential manipulation. This illustrates both the wisdom of crowds (through price discovery and information aggregation) and their madness (via noise, momentum, and participant bias), suggesting these markets are best used as complements to, rather than replacements for, expert analysis.
  • Read More
• ChatGPT is Transforming Peer Review — How Can We Use it Responsibly? - Nature
  • Analysis of 50,000 computer science conference peer reviews reveals that 7-17% now contain AI-generated content, with detectable patterns including formal tone, verbosity, and superficial analysis. While LLMs could help address reviewer shortages, they must be restricted to specific tasks like grammar correction and information retrieval, with clear guidelines needed to prevent undermining scientific integrity.
  • Read More
• Selection Effects - The Diff
  • Byrne Hobart examines the pervasive influence of selection effects, where observations are shaped not by reality but by the filters we encounter—media focusing on outliers, biased personal feedback, and self-reinforcing peer groups. Understanding selection effects can refine decision-making, but it also reveals a world shaped more by biases than universal truths, challenging us to question our assumptions while navigating our chosen bubbles.
  • Read More

Finance and Economics

• All the Devils ($) - Net Interest
  • Credit Suisse’s collapse, as chronicled by Duncan Mavin, was decades in the making—a result of mounting scandals, cost-cutting at the expense of risk controls, and cultural friction between Swiss conservatism and Wall Street ambition. Ultimately, the bank’s failure to align with the shifting post-9/11 regulatory landscape, especially under U.S. scrutiny, sealed its fate, highlighting the risks of entrenched dysfunction and the cost of ignoring strategic foresight.
  • Read More
• Misaligned Incentives - Clouded Judgement
  • As venture funds swell in size, GP incentives are shifting from maximizing long-term returns (20% carry) to prioritizing quick capital deployment (2% management fees). This trend often results in inflated valuations and large, early rounds, exposing founders to risks of misalignment, abandonment, or ‘deployment-first’ decisions. For founders, the crucial question becomes: are you partnering with a ‘2% firm’ focused on volume or a ‘20% firm’ committed to company outcomes?
  • Read More

Personal Development

• Maximizing Time for Reading - Blake Butler
  • Moving beyond productivity mantras, Butler presents reading as both liberation technology and neural training program, found in life's quiet margins and working through immersion rather than extraction. His essential insight: literature's apparent inefficiency is its hidden strength—it expands minds not through quick takeaways but through the gentle accumulation of experience, like a slow-release technology for consciousness.
  • Read More