Gianfranco's Best of January 2025 Reading List

Welcome to the January 2025 edition of my monthly reading list.  

This month, I've curated my favorite essays offering insights into the latest developments in AI, societal impacts of technology, strategic business thinking, financial trends, and personal growth.

What if the greatest innovations aren’t fueled by freedom—but by friction?

This month’s essays dissect how constraints—resource scarcity, temporal fragmentation, and systemic adversity—are not roadblocks but accelerants of ingenuity. Below, three pieces that reframe limitations as the invisible architects of disruption.

• DeepSeek Debates: Chinese Leadership On Cost, True Training Cost, Closed Model Margin Impacts (Dylan Patel, AJ Kourabi, Doug O'Laughlin, Reyk Knuhtsen)

  DeepSeek’s GPT-4-level models cost 90% less than Western peers—a feat achieved not despite China’s tech isolation but because of it. By pioneering breakthroughs like Multi-Head Latent Attention under rigid export controls, they’ve turned geopolitical constraints into a blueprint for hyper-efficiency. Yet as Western labs chase raw scale, it begs the question: does abundance dull our edge for innovation?

• Making Markets in Time (Abraham Thomas)

  Venture capital doesn’t just fund companies—it funds time slices. Thomas reveals how slicing startups into successive rounds transforms raw uncertainty into quantifiable, tradable risk units. But this temporal arbitrage has a dark side: optimizing for “consensus milestones” can breed incrementalism. When capital commoditizes time, who still bets on timeless ideas?

• HANDLE HARD WELL (Ted Lamade)

  Josh Allen wasn’t drafted to lead the NFL—until he weaponized rejection. Lamade parallels underdog athletes and resilient investors who thrive in volatility, arguing that adversity isn’t a barrier but a filter. In an era of easy capital and softer pivots, it poses a challenge: what if “talent” is the willingness to endure what others deem unbearable?

Connecting Theme: Constraints Create Catalysts
Each essay reveals how limitations propel meaningful progress—China’s AI leadership thrives on regulatory hurdles, venture capital’s time-slicing sharpens focus, and underdogs flourish when nothing is guaranteed. Far from stifling growth, constraints can become the very crucible of innovation.

Artificial Intelligence

• AI’s Uneven Arrival ($) - Stratechery by Ben Thompson
  • Ben Thompson argues that while advanced “inference-time scaling” models like o3 will enable AI to take on tasks independently—becoming “ammunition” rather than mere assistive “tools”—enterprises won’t necessarily seize these benefits quickly. He likens traditional companies to CPG advertisers who struggled to integrate targeted digital ads, suggesting organizational inertia will slow AI adoption beyond narrow job replacement. Thompson contends that truly transformative AI use will first unfold among new, more flexible businesses designed around agents, leaving established firms—and their seat-based SaaS model—grappling with a future that arrives at the periphery, then radiates inward.
  • Read More
• DeepSeek Debates: Chinese Leadership on Cost, Training True Cost, H100 Pricing Surge ($) - SemiAnalysis
  • SemiAnalysis argues the hype around DeepSeek’s “$6M training” figure ignores that R&D, data prep, and hardware TCO dwarf that number, reflecting deeper Chinese investment and new algorithmic breakthroughs. Meanwhile, MoE-based architectures and multi-head latent attention (MLA) reduce inference memory costs, enabling cheaper reasoning—yet also prompting broader questions about NVIDIA’s margins, regulatory constraints, and potential new export controls. Despite big leaps by DeepSeek, critics point out that Google’s and other labs’ models match or exceed this performance, and the real race is to integrate these new reasoning models profitably in a fast-evolving, geopolitically shaped AI environment.
  • Read More
• o1 isn’t a chat model (and that’s the point) - Latent Space
  • This guest post examines how OpenAI’s o1 model represents a shift from traditional chat-style interactions to a “report generator” approach, requiring extensive input and clear objectives for best results. It highlights o1’s ability to produce in-depth, accurate output on complex tasks (from coding to medical insights), while acknowledging its shortcomings in speed and stylistic output—ultimately challenging users and developers to rethink AI product design.
  • Read More
• Are Better Models Always Better? - Benedict Evans
  • Benedict Evans examines how “better” generative AI often just increases approximate correctness, rather than guaranteeing definitive, factual answers. Tasks like creative writing or coding can tolerate minor errors, so better models shine—yet for purely “right vs. wrong” requests (e.g. precise data lookup), probabilistic LLMs still fall short. He suggests new approaches and frameworks may emerge that harness the utility of approximation, rather than insisting LLMs produce strictly correct results in all contexts.
  • Read More
• 2025 AI & Semiconductor Outlook ($) - Doug O’Laughlin and Alpha Hunter
  • Doug O’Laughlin and Alpha Hunter argue that although AI-related stocks soared in 2024, the broader semiconductor industry had a difficult year, with automotive, analog, and smartphone segments especially weak. They emphasize that capital-intensive AI adoption is reshaping the technology landscape, urging investors to prepare for rising marginal costs and cyclical shifts—while still finding select “defensive” and long-term opportunities in memory, EDA, and AI hardware players.
  • Read More
• Capital Cycles and AI ($) - Doug O’Laughlin
  • Doug O’Laughlin examines the “capital cycle” framework—how investment rushes into profitable new technologies, overshoots, and eventually corrects—and applies it to today’s AI boom. He argues that AI’s compute and power demands are setting off a far-reaching cycle of infrastructure overbuild, drawing parallels to historical tech bubbles like railroads and telecom while noting key differences, such as the longer timelines for ramping up data centers and power.
  • Read More
• Things we learned about LLMs in 2024 - Simon Willison
  • Simon Willison reflects on a year of rapid advancement in large language models, highlighting the comprehensive “GPT-4 barrier” being broken by numerous competitors, dramatic cost drops driven by efficiency gains, and widespread adoption of multimodality (images, audio, video). He also flags the uneven state of LLM usability—where best practices, prompt engineering, and reliable tooling remain elusive—and examines emerging trends like inference-scaling “reasoning” models and the growing role of synthetic training data.
  • Read More
• DeepSeek: Is this Jevon's Cope? ($) - Doug O’Laughlin
  • Doug O’Laughlin dissects the market’s whiplash over AI infrastructure demand, arguing that while algorithmic breakthroughs like DeepSeek’s models spark fears of short-term GPU obsolescence, Jevons Paradox—where cheaper intelligence fuels explosive demand—will ultimately prevail. He observes a fragile market psyche, noting how “compute is still a huge bottleneck, and letting it up a little bit is good,” yet warns of sentiment shifts driven by pricing volatility and geopolitical tensions.
  • Read More
• Common Pitfalls When Building Generative AI Applications (Chip Huyen) - Chip Huyen
  • Chip Huyen highlights six recurring mistakes—like forcing generative AI on problems that don’t need it, ignoring “product over AI,” and overestimating early demos—which can derail real adoption. She warns that hasty complexity and little human oversight mask deeper user needs, ultimately denying teams the kind of “meaningful mastery” they aim for. By anchoring on sound product principles, steady evaluation, and strategic alignment, she argues, organizations can avoid funneling time into ill-fitting solutions and maximize genuine impact.
  • Read More

Finance and Economics

• Making Markets in Time (Pivotal by Abraham Thomas) - Abraham Thomas
  • Abraham Thomas explores how Silicon Valley finances high-risk innovation by “temporal arbitrage,” breaking a startup’s journey into rounds (seed, Series A, etc.) so each new investor merely must validate the next stage rather than the entire path. He likens this to market-making in commodity futures—where dividing and defining steps (when, where, how) enable liquidity and scale. Thomas argues that modern venture’s obsession with round benchmarks, consensus, and brand is precisely how it transforms a hazy future into legible, tradable slices of risk. By engineering time into stages and channeling massive capital inflows, large “venture majors” behave more like investment banks than old-school, value-seeking VCs.
  • Read More
• On Bubble Watch - Howard Marks
  • Howard Marks revisits the highs and lows of bubble psychology 25 years after his seminal memo “bubble.com,” noting how “no price too high” euphoria and fear of missing out have repeatedly driven stocks to unsustainable levels. He stresses that while major innovations can spawn genuine enthusiasm, market participants often overestimate persistence and underestimate cyclical turnarounds, leading to dangerous overpricing. In observing today’s lofty valuations—especially around tech leaders—Marks argues that a sober appreciation of history and investor psychology is key to avoiding the pitfalls of mania.
  • Read More
• A Personal Note From Our Founder (Hindenburg Research) - Nate Anderson
  • In announcing Hindenburg’s disbanding, Nate Anderson shares a deeply personal reflection: he describes realizing that a “successful career becomes a selfish act” once it overshadows other parts of life. Driven by self-doubt and needing to prove himself, he channeled intense dedication into investigative work—yet that singular focus often shut out loved ones. Now, having found a rare sense of self-acceptance, he’s ready to step away, viewing Hindenburg as a chapter in his life rather than his defining purpose.
  • Read More
• The Most Powerful Tax Break in America (Silly Money) - Ankur Nagpal
  • Ankur Nagpal unveils QSBS, calling it “the single most generous tax break in America,” a lifeline for founders, employees, and investors who seek more than fleeting wins. By protecting and even multiplying untaxed gains—through thoughtful share structures, gifting, and trust setups—he reframes startup success as something sustainable that can align with deeper ambitions beyond a five-year horizon.
  • Read More

Science and Society

• Learn smart lessons from the L.A. fires, not stupid lessons ($) - Noah Smith
  • Noah Smith warns against politicized misinformation surrounding the 2025 L.A. wildfires, stressing that climate change has created a new era of heightened wildfire risk demanding pragmatic adaptation. He critiques California’s restrictive insurance policies and environmental regulations that hamper proactive forest management, urging more controlled burns, fire-resistant building practices, and realistic approaches to insurance as part of long-term disaster preparedness.
  • Read More
• TikTok is just the beginning ($) - Noah Smith
  • Noah Smith argues that China’s government is determined to weaken or control foreign societies, and that TikTok — already found to suppress anti-CCP content — is simply one tool among many. He frames the possible U.S. ban or forced sale of TikTok as emblematic of a deeper clash between liberal democracies and an increasingly assertive, tech-enabled authoritarian power. Whether the West can summon the will and capacity to defend against these new forms of state-backed influence will be a key test of liberalism’s resilience.
  • Read More
• Live lindy in 2025 - Adam Singer
  • Adam Singer advocates returning to time-tested practices, from unprocessed foods and daily physical movement to remote work and natural sleep schedules, as a way to combat modern malaise. He suggests that adopting “lindy” behaviors—habits shaped by evolution—can foster better health, deeper community engagement, and personal fulfillment in an era overwhelmed by novelty and distraction.
  • Read More

Personal Growth and Learning

• The Best Tacit Knowledge Videos on Every Subject - LessWrong
  • This curated post assembles a wide-ranging list of “tacit knowledge” videos—resources that convey practical, hands-on expertise often difficult to transmit through text alone. It covers diverse fields such as software engineering, business communication, and cooking, offering examples where experts demonstrate real-world processes, methods, and decision-making in a transparent, detailed way.
  • Read More
• An Unreasonable Amount of Time - Allen Pike
  • Allen Pike illustrates how Teller’s “buried card” trick—prepared months in advance—is a testament to spending far more time on a goal than anyone would imagine. He parallels this with other crafts, noting that extraordinary achievements often stem from consistent, extensive effort that creates an air of “magic” for onlookers.
  • Read More
• HANDLE HARD WELL - Ted Lamade
  • Ted Lamade draws a contrast between five-star recruits—touted from day one but often complacent—and those labeled only three-stars, forced to contend with real work, late physical growth, and myriad sports. The underdogs’ lack of early spotlight became precisely the ingredient that sharpened their grit, so that once opportunity struck, they’d already mastered the hustle. It’s a reminder for anyone chasing success—that being overlooked can serve as the ultimate edge, especially in a climate where easy confidence crumbles and those who’ve toiled quietly, “blooming where they’re planted,” stand ready to shine.
  • Read More

Web3

• Up or Down? Lessons Learned and Forgotten - Matti
  • Matti revisits the prior bull cycle’s euphoria—fueled by novel DeFi tech—and contrasts it with today’s environment, where speculative memecoins, fleeting narratives, and top-down factors like regulatory shifts rule the market. She argues that the current hype lacks a true innovation trigger, leaving many hopeful for an imminent “dotcom-scale” mania that might never come. In this cautionary lull, crypto users and builders question whether any broad-based upswing is near or if the entire sector is drifting toward disillusionment.
  • Read More
• The Fat App Thesis - David Phelps
  • David Phelps contends that while protocols have historically dominated crypto valuations, the real long-term upside now resides in applications. He argues that user-oriented apps—with unique financial and social features made possible by onchain infrastructure—will generate far greater value and ultimately power the blockchains underneath them.
  • Read More
• Stablecoins in 1,000 words - Matt Brown
  • Matt Brown compares stablecoins to retrofitting reliable, fiat-backed “carriages” onto the fast, global rails of blockchain: they