AI

Reading guide

This post reprints an essay about the narrative of “machines replacing humans.” In Homo Deus, Harari proposes that the most important economic question of the twenty-first century is how to create jobs for “useless people,” envisioning a small elite using technology to become “Homo deus.” The essay says this narrative dates to Aristotle, while the Industrial Revolution shows that machines both destroy and create jobs. The possible difference this time is that earlier machines replaced physical labor, whereas future intelligent machines aim to take over mental work across the board. Robin ends with a question: once printing became widespread, copying manuscripts by hand disappeared, but humanity moved to another stage. Are humans now being replaced, or are we on the eve of a new chapter? Harari’s predictions are relayed by the essay as hypothetical scenarios, not facts.

Reading guide

An examination of VC behavior through the influence people have on one another in finance: the industry has moved from an era of lone wolves to an era of the herd, with rotating hot sectors and increasingly similar investments. The argument favors independent judgment and perspectives from other disciplines as ways to look beyond consensus. Cautions: the “era of the herd” is an observation, not a statistical finding. Disagreeing with consensus does not make you right, and a cross-disciplinary perspective cannot replace professional due diligence. Independent thinking does not rule out exchanging ideas.

Reading guide

Robin's late-2022 view of startup fundraising over the next one to two years: pandemic-driven remote demand and quantitative easing fueled capital markets and startup investment in 2020–2021. In 2022, the continuing Russia–Ukraine war, inflation following a decade of easy money, tightening and rate hikes in Europe and the United States, and spending shifting from electronics to offline activities battered technology stocks. Meta and Netflix fell over 70%; AMD, NVIDIA, and Intel over 60%; Zoom and C3.ai as much as 90%. Sequoia and Y Combinator warned of a funding winter. Startups must return to fundamentals: will customers pay for the product or service, rather than merely hoping VCs will? The era of burning cash for market share has passed. Be more focused and lean, and right-size the organization if necessary to weather the next one to two years. VCs assess investments over five to seven years, rather than being entirely driven by two to three years of volatility, but large portfolio write-downs create operating pressure and may require valuation adjustments. Stock declines are figures from 2022.

Reading guide

Robin opens with his view that the hottest investment area of 2022 has returned to AI, especially the widely discussed topic of AI generation: the future is here. The reprinted 36Kr Shenyiju report describes a San Francisco generative-AI gathering organized by Conviction founder Sarah Guo, bringing together engineers, entrepreneurs, and financiers, including Hugging Face's CEO. Stability AI announced $101 million in funding at a $1 billion valuation; Jasper raised $125 million at a $1.5 billion valuation. Microsoft's Copilot and OpenAI's GPT-3 and DALL-E 2 fueled the excitement, with even Sergey Brin appearing at a related gathering. Investor Nathan Benaich observed considerable FOMO. Legal questions were emerging too: the boundaries of fair use for training models on copyrighted data were expected to be tested in court. Valuations and funding amounts are figures from 2022.

Reading guide

This excerpt from the Chinese edition of The Power Law, translated by Professor Xuan Tian, describes VC as a bridge between dreams and reality, once called humanity's last hope by The Wall Street Journal. VCs find promising seeds among imaginative ideas, nurture them, and eventually hand them to more suitable owners before stepping back. The industry's pioneers, uncertainty, decline, and renewal intertwine with new challengers. Yet hidden corners remain: growth-driven governance failures and scandals at Uber, WeWork, Theranos, and FTX warn that our understanding is incomplete. Robin adds two comments. Taiwan's VC engine has stalled somewhat, but he is confident it can restart and accelerate to make up for more than a decade of stagnation. VC carries dreams: its few celebrated successes imply many failures, yet losses must not extinguish confidence or hope. Bruised investors carry on, learning from bad investments and missed opportunities. He closes by adapting a line from The Three-Body Problem, described as essential reading in China's VC community: “We will give civilization to time, rather than time to civilization.”

Reading guide

Robin argues that material information announcements on the Market Observation Post System should be factual and restrained. If a company wants to celebrate how wonderful it is, he suggests buying media space: Taiwanese readers and investors cannot distinguish a journalist's reporting from a company's promotional copy, and media outlets need advertising budgets to survive. Announcements on the exchange-supervised disclosure platform, however, must take the regulator's approach and standards into account. In the example, passive-component manufacturer YAGEO issued a September 26 announcement titled “YAGEO Group AI Summit Concludes Successfully, with Every Seat Filled.” On October 7, the Taiwan Stock Exchange imposed an NT$200,000 contractual penalty because material information announcements must not contain advertising-like statements. It reiterated that such disclosures must not provide unconfirmed information, expound management philosophy, or serve as an advertising platform.

Reading guide

Robin reviews a landmark acquisition: cloud data company Databricks announced a US$1.3 billion acquisition of MosaicML, an AI startup less than three years old. Founded in 2021 with just 62 employees, MosaicML had disclosed only one financing round, at a US$220 million valuation; the acquisition represented a sixfold jump. Founded by a group of former Intel executives, its team enables enterprises to train and deploy AI models in secure environments. Automated training optimization promises training speeds two to seven times those of traditional methods. Databricks plans to combine its Lakehouse platform with MosaicML so enterprises can train generative AI on proprietary data at low cost. Robin's judgment: technology giants' acquisitions of generative AI startups are only beginning. Figures reflect company statements from June 2023.

Reading guide

This reprint of Wu Bofan's weekly business commentary examines Amazon. There are no ready answers about the future: users do not state them directly, and only extraordinary attention reveals signs of latent needs. The depth of your understanding of needs determines how far your product can reach. In 26 years, Amazon experienced more than ten major failures and billions of dollars in direct losses, but its apparent hesitation was really iteration. Bezos emphasizes repeatedly taking detours through nonfatal failures to approach success. Capabilities accumulated through hardware failures later combined with new needs to underpin Alexa and its voice platform. Everyone understands “fear standing still, not moving slowly,” but without supporting culture and processes, principles become slogans. The article uses Parkinson's Law to explain organizational entropy: more people require more coordination, consuming resources without creating value. Bezos's Day One approach puts as many people as possible directly in front of customers, solving concrete problems with the obsession of the company's first day.

Reading guide

Robin explicitly argues that managers and employee referrals should bring in people stronger than themselves. The quoted material offers two classic stories. Ogilvy’s founder David Ogilvy used Russian nesting dolls to illustrate that hiring smaller people makes a company of dwarfs, while hiring stronger people makes a company of giants. Wu Jun describes Google’s early recruiting tradition: making room for better people and hiring stronger talent can help both individuals and organizations grow in Silicon Valley’s competitive environment. The quoted claims about competition in Silicon Valley are the original author’s views. Robin’s stated agreement concerns hiring stronger people and does not extend to every exclusionary argument in the quoted passage.

Reading guide

Lei Jun’s experience suggests that hundreds of bilingual business-plan pages can fail to secure investment. Busy investors begin with the recommender’s credibility, then assess the team, then give you five minutes to explain what you do and the opportunity five to ten years ahead. Robin draws four lessons: seek a reliable personal introduction from someone willing to invest too; do not scatter business plans indiscriminately; make the first three to five minutes count; and get to the point without fashionable jargon. Disliking long plans does not eliminate information needs, and a recommender’s willingness to invest is a trust signal rather than a substitute for due diligence.

Reading guide

In his New Year’s Day 2022 reflection, Robin observes that the pandemic changed life and work. Rather than chase every change, invest resources in enduring fundamentals. At work, portfolio company 91APP listed and became a Taiwan-born startup unicorn with market value above 30 billion; Great Tree’s revenue exceeded ten billion, and Zhen Yu Hardware listed on the OTC market. These were milestones, not endpoints. Robin sees VC itself as a traditional industry that must evolve; Taiwan’s VC, corporate VC, and private equity sectors need their own paths. He anticipates Web3, the metaverse, and digital assets shaping the next decade. Personally, he continues learning how to parent Shen Xiaobao. Values, revenue, and project progress reflect the end of 2021.

Reading guide

Robin reviews eight investment waves during pandemic-shadowed 2021: Clubhouse’s early surge and later silence, though audio-based knowledge sharing strengthened; electric vehicles ahead of carmakers’ combustion-engine deadlines; AI and robotics; space travel and low-Earth-orbit satellites, including Bezos, Branson, and Starlink; the metaverse, propelled by Facebook and NVIDIA; blockchain-based Web3 and its connection to metaverse narratives; biotechnology and gene editing; and quieter but continuing SaaS development. In a tenfold-speed era, he expects faster progress over the coming decade, with VC continuing to enable change. These judgments reflect late 2021, not current investment advice.

Reading guide

Robin’s personal judgment is that anyone choosing to invest in China should seek sectors the government currently supports, or at least does not strongly suppress. He lists internet services, fintech such as Ant, education, healthcare, property, entertainment, gaming, and livestreaming as targets of pressure. Supported fields tend to involve hardware or perceived social stability: semiconductor self-sufficiency, critical components, surveillance, AI and IoT smart cities, agricultural and food security, defense, aerospace and satellites, EVs, batteries, energy, environmental protection, and industrial automation. These are personal observations from May 2022, not investment advice.

Reading guide

Robin expresses admiration for AI company Kneron. MTS AI, a subsidiary of Russia’s largest telecom operator MTS, signed an agreement to invest around US$10 million in the Taiwanese edge AI chip developer. MTS AI would receive exclusive Russian distribution rights for Kneron chips and use its technology to develop an AI product line of its own. Robin notes that Taiwanese AI software had just delivered good news in Japan and hopes Taiwan’s hardware, an even greater source of pride, would bring good news from the U.S. market within a year or two. The investment and market developments refer to April 2021.

Reading guide

Robin recounts a hiring interview. A strong candidate—second-round VC interviewees generally have good backgrounds—was asked to recommend an AI investment. The candidate named Knxxxx, which Robin’s firm had also backed that year. The problem was not the company; it was recommending it simply after reading favorable coverage, without a distinctive view that could withstand deeper questions. Journalists and taxi drivers have abundant information too. Due diligence means using the product, assessing strengths and weaknesses, comparing the segment and competitors, and explaining a view of your own. Investing requires independent judgment. The company name remains masked as in the source.

Reading guide

Robin offers five observations on recovering startup relief loans. Administrators know many recipients probably cannot repay, but must follow written rules and contracts. Civil servants who depart from those contracts risk accusations of improper favoritism or dereliction of duty. The National Development Fund is public money, subject to auditors, accounting authorities, and legislators. Officials cannot simply initiate extensions or waivers, but an external body such as the legislature can convene discussion of general rules, giving lawful implementation more flexibility. Converting relief preferred shares into common shares is also difficult: what conversion price is fair, and could par-value or net-asset-value conversion make the fund a controlling shareholder and turn a private startup into a state enterprise? Robin also calls for more innovation, investment, and open capital markets in response to AI.

Reading guide

Andrew Chen’s newsletter explores the dark side of being data-driven. Data measures yesterday rather than tomorrow and can produce analysis paralysis: intuition may provide an answer in seconds, whereas A/B tests take weeks. The tyranny of the majority can reduce a product to a compromise around averages. Noise and false signals abound, correlation is not causation, and historical data struggles to predict a future that has never occurred. Startups competing with giants such as FAANG need intuition, speed, and bold moves rather than copying those companies’ data processes. The ideal is to be data-informed, neither blindly data-driven nor data-ignorant.

Reading guide

Cai Yu examines Hefei’s investment-based approach to attracting industry. When BOE considered locating there, Hefei backstopped RMB17.5 billion. When Sharp intervened, Hefei offered to accept both projects, prompting Sharp to withdraw. It also invested in LCD and plasma technologies to hedge competing technical paths. Investments in ChangXin, GigaDevice, and NIO followed. The RMB7 billion NIO investment involved a NIO China joint venture, a RMB420 billion revenue commitment, an IPO within 60 months, and clear exit arrangements. The underlying approach is to follow a coherent industrial strategy, buy promising assets when they are out of favor, and train investment-promotion teams to conduct due diligence like investment managers. Robin suggests that Kaohsiung, Taichung, and Taoyuan could learn from this ambition. A city should first answer four questions: Who am I? Where am I now? Where do I want to go? How should I get there?

Reading guide

A Caijing article discusses the “one-person unicorn” in the AI era. Li Zhifei used AI to produce 40,000 lines of code in two days; Sam Altman described “one person plus 10,000 GPUs.” Pokee AI, associated with Zhu Zheqing, raised a $12 million seed round with only six people. A view attributed to Sequoia suggests hierarchical organizations are giving way to dynamic collaboration. Highly capable generalists and lean AI-powered solo businesses are emerging, while Chen Yiran comments on the employment market. Cal AI, built by high-school students including 17-year-olds, reportedly earns $1.12 million a month. Robin reminds venture capital professionals that they, too, must keep pace with the times.

Reading guide

Robin recommends the 2016 shareholder letter as a must-read. Bezos describes Day 2 as stasis, followed by irrelevance, painful decline, and death: a company must remain in Day 1. Four essentials defend it. First, genuine customer obsession: customers are always “beautifully dissatisfied.” Second, resist proxies: processes and market research cannot replace closeness to customers. Third, embrace external trends—then, machine learning and AI. Fourth, make decisions quickly: act with about 70% of the information, disagree and commit, escalate misalignment promptly, and avoid one-size-fits-all processes.

Reading guide

Robin’s opening question is what kind of education we need in the AI era. Having raised related educational issues over the preceding few days, he turns to Wan Weigang’s perspective. Wan argues that traditional education takes a top-down approach: schools, teachers, and parents hold the initiative. Once likened to a gardener cultivating plants, it now has a more precise analogy—training AI. Defining the learning scope in advance, feeding in materials labeled right or wrong, and assessing the results is supervised learning. Yet students taught this way fall short even of GPT, which uses self-supervised learning, discovers patterns for itself, and is designed to know things its teachers do not. Traditional education and AI share three shortcomings that innovators do not: rewards depend on administrators’ approval, whereas innovators gain recognition from society and consumers, without fixed standards, and may even define what society considers good; mistakes and shoring up weaknesses receive great attention, whereas innovators need exceptional strengths; and new things outside the defined scope attract little interest, whereas innovators pursue novelty because it may confer a competitive advantage. The biggest problem with AI-style education is students’ lack of autonomy. A machine’s great advantage is its inherent passivity; a defining human difference is the desire to take the initiative. The conclusion: in an age with AI, we must think carefully about nurturing a person rather than training one. Creative people can be nurtured but cannot be controlled. To hope for unexpected good, we must tolerate unexpected “bad”—things we did not want them to do. AI can be controlled; with talented people, we must learn to tolerate their independence.

Reading guide

Robin argues that income polarization in Taiwan will become increasingly pronounced. In the future, 50–70% of people may pay no income tax, while the top 30%, or even the top 5% or 10%, could support the entire income-tax contribution. Fewer taxpayers would not necessarily mean declining revenue: the top 30% remain on a high-growth trajectory, earning more and therefore paying more. Even as taxpayer numbers shrink, collections can exceed government budget estimates, with most of the contribution coming from a small group at the top. The post gives this distribution: 0.13% of people pay 17.47% of income tax; 0.85% pay 41.19%; 2.79% pay 60.75%; and 7.66% pay 77.21%. A small group, around 10% of Taiwan’s population, benefits from globalization, AI development, and U.S.–China confrontation. They earn “Global Pay” and can expect further income growth. A much larger group earns “Local Pay,” primarily in domestic services or traditional industries competing intensely with China. Their income growth may be limited, stagnant, or even persistently negative. As high earners drive inflation and increases in goods and housing prices, Robin argues that government needs tax and subsidy measures to support disadvantaged groups left behind by globalization and AI. They may feel disconnected from the past decade’s economic growth while experiencing its inflation keenly, creating a sense of relative deprivation.

Reading guide

Robin observes that Taiwan’s exports have boomed while domestic demand has remained weak over the past three to five years, rapidly widening the wealth gap. Over the preceding five years, the world experienced a once-in-a-century pandemic, an AI boom, and U.S.–China confrontation. He argues that Taiwan ranks among the top three beneficiaries: the pandemic drove demand for electronics used remotely; AI generated overwhelming orders for semiconductors and servers, outpacing factory expansion; and U.S.–China tensions shifted electronics orders from China to Taiwan. Export-related technology stocks, including semiconductors, servers, and components such as thermal-management products, surged, and people in those industries earned handsomely. Domestic demand faced pressure from weak tourism, labor shortages and rising wages, and inflation including higher utility costs. Traditional manufacturers such as machinery companies also faced intense competition. A minority riding these trends saw incomes soar and pushed up housing and goods prices, while the majority outside them faced greater economic pressure. Robin adds that declining birth rates are, in his personal view, the larger and more irreversible crisis.

Reading guide

Robin shares statistics identified as being from November 2024. After the Taiwan Stock Exchange’s market capitalization exceeded 90 trillion, its listed companies’ combined value had reached eighth or ninth globally, roughly equal to Germany’s stock market. The comparison gives Germany’s GDP as about US$4.31 trillion, with 85 million people and US$52,000 per person, and Taiwan’s GDP as about US$930 billion, with 23 million people and US$41,000 per person. Germany’s GDP was 4.6 times Taiwan’s, but their stock-market values were already comparable, even excluding Taiwan’s Taipei Exchange. Returning to an Economist article, Robin argues that Taiwan benefited from the preceding five years’ AI boom: outside the United States, its increase in stock-market capitalization was second only to that of Nasdaq, the market at the origin of the AI boom. In that sense, he considers it reasonable to call Taiwan the greatest beneficiary outside the United States, while noting that the gains were concentrated in its core competitive strengths.

Reading guide

Robin observes that TNMG, an Asia-focused next-generation digital technology and media group listed on Nasdaq, had reached a market capitalization of US$5.6 million. Its businesses span media, AI, traffic, data analytics, advertising technology, ecommerce, and B2B. It attracts more than 45 million unique users each month, with Japan its largest market and Taiwan second. Yet with 2024 revenue of US$48.5 million, its price-to-sales ratio was only 0.12. Robin poses two questions: might a Taiwan listing offer the chance of a tenfold market valuation? And does the group need such a wide range of businesses, or would it be better to focus on one or two and make them globally competitive?

Reading guide

Robin cites The Economist’s observation that Taiwan benefited from the preceding five years’ AI boom, with its stock-market capitalization increasing at a rate second only to Nasdaq in the United States, the origin of the boom. He adds that it is reasonable to call Taiwan the largest beneficiary outside the United States, while noting that the benefit is highly concentrated in the foundations of Taiwan’s core competitive strengths. The figures are quotations reflecting their original time, not current facts.

Reading guide

Robin identifies a blind spot common among technical founders: carrying a hammer and looking everywhere for nails. Blockchain startups a few years ago, and AI startups in more recent years, are the clearest examples. Both can be seen as taking a technology and looking everywhere for possible use cases.

Reading guide

The post observes that Taiwan has roughly 23 million people—less than 0.3% of the world’s population—yet has a profound impact in the AI era. As a small country with limited resources, especially human resources, Taiwan should focus on fields where it can compete and win. Facing the challenges of a new era, it needs strategy, focus, and trade-offs so it can become one of the crucial few. It need not expect to win an Olympic gold medal in every event. Even South Korea and China focus on events where they can win medals, while firmly defending the disciplines in which they excel.

Reading guide

Robin proposes the direction “AI + Edge Computing + Smart XX,” where XX could be manufacturing, healthcare, factories, retail, cities, homes, agriculture, transportation, and more. He wants to organize an investment framework around cloud, connectivity, edge devices, and applications: cloud includes cloud services, streaming, and AI; connectivity includes 5G/6G and low-Earth-orbit satellites, which he marks with a question; the edge means edge computing. His rationale is that an autonomous vehicle cannot wait to upload images to the cloud and receive a response. Hospitals, factories, and households may also not want to upload every piece of data. He hopes investors can capture the opportunities presented by vast numbers of endpoints in the AIoT era.

Reading guide

Robin recalls that in his years in the industry, the vocabulary has moved from IoT, big data, blockchain, and VR/AR/MR/XR to AI in more recent years, and now—2021—the metaverse. He cautions against dressing things up with too much jargon. Explain clearly and simply what you actually want to do and which problem you want to solve; glittering terminology can conceal empty substance. He also observes that the blockchain community has cycled through ICOs, STOs, IEOs, DeFi, and NFTs. It often feels like one term replacing another, and he is unsure whether much has changed at the fundamental level. These are his personal observations.

Reading guide

The post explains that AIGC stands for AI-generated content: artificial intelligence can automatically generate news articles, interactive conversations, images, speech, music, code, legal documents, financial statements, and more. It ends by suggesting that when AI becomes proficient in playing instruments, board games, calligraphy, painting, poetry, prose, music, and office work—and can also drive itself—humanity must consider its own value at a more philosophical level. The source and originality remain unconfirmed. This is a conceptual explanation from December 2022.

Reading guide

The post draws on Harari’s argument in Sapiens: humans can build large-scale social organizations because they have the extraordinary ability to imagine things that do not exist. Shared imagination makes cooperation at scale possible. Robin’s personal view is that AI still lacks imagination, including the ability to derive new knowledge and generate new insights. It functions more like a system that assembles and presents large amounts of existing information, and it may even lack the ability to make value judgments. For example, when asked whether vaccines are beneficial or harmful, AI can gather much of the information online and draw a conclusion, but that conclusion still rests on existing scientific data available online. This is a personal assessment from December 2022; AI’s capabilities have changed since then.

Reading guide

The post retells an analogy: when an apple hits Newton, he abstracts a theory of universal gravitation; when it hits ChatGPT, it merely explains what happens when an apple hits you, how to avoid it, and the benefits of eating apples. The ability to connect seemingly unrelated things and develop a systematic theory remains, at that time, ChatGPT’s ceiling. Robin adds that a Newton-level genius may appear only once among millions of people. Most people’s routine work still faces considerable risk in the foreseeable future. If someone else can do your job just as well, the risk is high. In his example, A actively embraces ChatGPT and may become so much more productive that A replaces B, who passively waits.

Reading guide

The post argues for starting with a change in how we teach. Traditional large classes can become one-on-one, tutor-style question-and-answer exchanges in the AI era. Taiwanese students, who have not been accustomed to asking questions, must now actively develop and organize their own questions. In the past, the worry was that a teacher could not answer. Now the worry is that we cannot formulate questions well: asking the right question matters more than being given an answer directly. If an explanation is unclear, students can keep asking, request a different explanation, ask for language an elementary school student can understand, or seek more depth and examples. The author argues that education in this new era needs to be redesigned.

Reading guide

Robin sees a critical moment for transforming education: Taiwanese children spend their days on workbooks and complete more than ten thousand test papers, while American children work on hundreds of projects from an early age, developing teamwork and problem-solving skills. The excerpt that follows comes from Wan Weigang’s column. It argues that education needs to change even without the AI wave; AI simply makes the problems more visible. In programming, for example, the question should not be which language you studied or how many facts you mastered, but which projects you have completed. We need to move from a “course mindset” to a “project mindset.” The excerpt also argues that GPT helps people learn faster rather than giving them a reason not to learn.

Reading guide

The post first quotes Wang Xing: a company’s starting point is not watching competitors but focusing on customers and creating value for them. A company can exist only if customers are willing to pay. Robin then adds that the internet has produced many new services, and now the AI moment has arrived. He expects many existing workflows to be disrupted or reinvented. Services that once required hundreds or thousands of people may become possible with a few dozen or ten-plus people. They are worth reimagining and doing over again. This is Robin’s expectation as of May 2023.

Reading guide

Robin notes that NVIDIA’s Taiwanese partners include both major hardware companies and nascent software startups. Amid the AI wave, the transition from old to new, and hardware–software integration, he believes Taiwan can play more critical roles. He cites NVIDIA’s market capitalization that day as approximately US$783.4 billion, the fifth largest in the U.S. market; these figures are as of May 19, 2023, and have not been independently verified. He hopes this will help Taiwan keep upgrading and transforming in AI. He ends with a retrospective quip: investing several years earlier, when the market capitalization was around US$20 billion, would have produced roughly a fortyfold return over seven years. This is not investment advice.

Reading guide

The post reprints a Dedao Headlines piece explaining why some powerful technology companies have so few people. It uses Midjourney, with approximately eleven employees, as an example; the figures are those stated in the reprint as of June 2023 and have not been independently verified. The analysis offers three reasons: low marginal costs in technology product development, mature infrastructure that removes the need to reinvent the wheel, and the 10x engineer effect. An additional section discusses team management at the “critical point of a phase transition.” Robin’s closing comment is that if revenue growth is highly correlated with headcount growth, the business is fundamentally a traditional, labor-driven enterprise rather than a technology company.

Reading guide

The post reprints a Wan Weigang article, tagged #WanWeigangEliteDailyLessons5, arguing that entrepreneurs’ most important decisions rest on intuition and imagination rather than scientific rationality—for example, building rockets or betting on large language models. It cites Zhang Weiying’s Chongxin Lijie Qiyejia Jingshen (“Rethinking Entrepreneurship”): someone else saying you cannot succeed does not mean you actually cannot. The author says execution requires science and reason, but an entrepreneur’s role is to initiate a vision beyond normal operations and discover new possibilities. The world contains not only quantifiable risk but also unquantifiable uncertainty—Knightian uncertainty. The entrepreneur’s task is to create uncertainty rather than adapt to risk.

Reading guide

The excerpt describes the cycle of enthusiasm for new technology: it is initially hyped, expectations then fall to a trough, but technology that genuinely advances digitalization and intelligence can eventually surpass those original expectations. It describes AI as overheated at the time and argues that extraordinary entrepreneurial revolutions often arise outside the consensus as markets cycle between excessive enthusiasm and excessive pessimism. Robin adds that investors can wait and see, but founders cannot remain on the sidelines. They must keep moving and adjust flexibly along the way. He also urges caution about market consensus: good investors and entrepreneurs need independent views.