第 1 章
When Machines Think Like Us: A Journey into the Future of AI
Imagine a world where your digital assistant doesn't just respond to commands but anticipates your needs, writes your emails with your distinctive voice, and helps solve your most complex problems. This isn't science fiction-it's happening now. When Reid Hoffman, LinkedIn co-founder and tech visionary, first interacted with GPT-4, he experienced what he calls an "AHA!" moment-Amplifying Human Abilities. Unlike previous AI systems that produced mediocre responses, GPT-4 demonstrated unprecedented contextual understanding and adaptability. The technology quickly caught the attention of industry leaders and celebrities alike, with figures from Elon Musk to Barack Obama acknowledging its transformative potential. In "Impromptu," Hoffman collaborates with GPT-4 itself to explore how this technology is reshaping our world-not as a comprehensive technical manual but as a traveler's guide to a new frontier where human creativity and machine intelligence converge to create something greater than either could alone.
第 2 章
The Education Revolution: AI as Teacher, Tutor, and Tool
When seventy-year-old Professor Steven Mintz first encountered ChatGPT, he didn't panic about academic integrity or ban its use. Instead, this veteran educator with perfect student ratings immediately recognized its potential as a teaching tool. Within months, he required his University of Texas seminar students to write essays collaboratively with AI, bringing their prompts and the machine's responses for class discussion. For Mintz, ChatGPT represents the latest in a series of technologies that have transformed education-like Google, calculators, and Wikipedia before it.
"If machines can do certain tasks well," Mintz argues, "humans shouldn't duplicate those abilities but surpass them." Drawing on organizational psychologist Tomas Chamorro-Premuzic's work, he suggests humans can thrive alongside AI by focusing on uniquely human capabilities: asking the best questions, learning insights not available in AI training data, and turning insights into actions.
This perspective is gaining traction among forward-thinking educators. Cherie Shields, a veteran public school English teacher, emerged as an early AI advocate, arguing in Education Week that rather than banning ChatGPT, teachers should embrace its potential to improve student writing. In a New York Times interview, she reframed concerns about "cheating" as "assistance" and demonstrated how ChatGPT could provide immediate feedback on student essays-a transformative capability when teachers typically grade 80+ essays weekly.
The stakes couldn't be higher. The divergent paths of two men named Wes Moore illustrates education's profound impact. Governor Westley Moore, whose family sacrificed everything for quality private schooling, became a Rhodes Scholar and Maryland's governor. Meanwhile, Wesley J. Moore, born in similar circumstances but attending underperforming public schools, received a life sentence for murder. Their stories raise fundamental questions about education's power to overcome socioeconomic determinism.
Despite America's historical expectation that schools deliver salvation for students and society, outcomes remain largely determined by parental income and zip code. Neither technology alone nor increased funding has significantly improved outcomes for low-income students. However, AI could transform teaching by providing immediate personalized feedback, creating customized assessments, designing personalized learning paths, facilitating collaborative experiences, and moderating interactive discussions.
The potential impact extends globally. With 40% of Africa's population under fifteen (560 million children) and 60% not attending school by age fifteen, educational challenges are immense. Bridge International Academies demonstrated remarkable results with tablet-based standardized lessons, helping students gain three years of learning in just two years according to Nobel laureate Michael Kremer's studies. Though Bridge's business model ultimately failed, combining LLMs with tablet-based instruction could help hundreds of millions of children worldwide who currently receive little or no education.
第 3 章
Creative Partnership: Human Imagination Meets Machine Intelligence
When Reid shared AI's creative potential with a Grammy-winning musician, the artist was initially terrified by the prospect of AI creating music in John Lennon's style. However, fear quickly turned to excitement upon realizing how AI could serve as a collaborative tool-generating multiple creative options that he could select from, modify, and enhance with his own artistic vision.
This collaboration between human creativity and machine capability represents both opportunity and challenge. On the positive side, AI offers new tools for expression, helping artists explore genres, styles, and techniques that might otherwise be inaccessible. However, AI also threatens creative authenticity and autonomy. Artists might face plagiarism accusations if AI generates content similar to existing works, lose their distinctive style through overreliance on AI, or struggle to maintain their unique voice.
Perhaps most concerning is AI's potential to flood creative markets with generated content, making it harder for human artists to stand out and earn income. Despite these challenges, AI will likely become an essential creative tool-somewhere between highly capable assistant and creative partner.
The creative community is engaging with AI at an accelerating pace. GPT-based tools can enhance various workflows: videogame designers can generate engaging dialogue for interactive characters; architects can create 3D models from sketches; and sculptors can analyze existing works to propose new variations. Though some applications remain hypothetical, GPT is poised to become the next major technology platform-potentially bigger than the internet-with applications developing rapidly.
Despite this excitement, legitimate concerns remain about potential negative effects: cultural homogenization through mass-produced content lacking originality; manipulation through deceptive content; and displacement of human creators. Mitigating these risks requires ethical standards, consumer education, and economic support for human creators.
The relationship between creators and AI will likely remain dynamic and sometimes contentious, as demonstrated by tools like Glaze that help artists protect their styles from being mimicked by AI systems. The best approach is to engage critically and creatively with AI rather than passively accepting its output.
第 4 章
Justice Reimagined: AI's Role in Creating Fairer Systems
Justice systems are a critical measure of society's moral character, particularly in how they treat the poor and powerless. The Legacy Museum in Montgomery reminds us of slavery's lasting impact on American justice, with Michelle Alexander's work revealing that more Black men are involved in the carceral system today than were enslaved in 1850. This gross injustice demands attention, and GPT-4 could potentially help address these systemic problems.
The U.S. justice system has one of the world's highest incarceration rates. Since AI is already being used in justice systems and won't disappear, we must actively engage in shaping how these technologies are applied rather than allowing institutions that created structural inequalities to define their use.
AI in criminal justice has already demonstrated serious problems. Predictive policing systems have embedded human biases that are then legitimized by the algorithm's apparent objectivity. Facial recognition algorithms have shown higher false positive rates for darker-skinned individuals. Rather than focusing solely on these downsides, GPT-4's design, which puts AI power into individual citizens' hands, could democratically empower those traditionally marginalized by state power.
Body-worn cameras (BWCs) are becoming increasingly prevalent in U.S. law enforcement, with nearly half of agencies using them and up to 80% in large departments. While they provide crucial oversight for police violence issues, they risk becoming surveillance tools without proper policies. AI can enhance BWC benefits by automatically redacting sensitive information, identifying key moments in footage, detecting misconduct, and analyzing patterns to improve training. Companies like JusticeText are using AI to help public defenders analyze video evidence, automatically transcribing and tagging data to ensure defendants have equal access to evidence as prosecutors.
GPT-4 and similar AI systems could significantly improve legal representation for those who cannot afford expensive attorneys. As a copilot for overworked public defenders, AI could quickly search legal databases, recommend strategies, and make complex legal matters more accessible to those unfamiliar with the system.
Prison populations face severe limitations on internet access, high illiteracy rates (75% of U.S. prisoners are illiterate), and isolation that breeds negative outcomes. AI could counter these issues by providing educational materials to improve literacy, offering virtual mentoring to develop social skills and career pathways, and connecting inmates with supportive communities.
White-collar crime costs the U.S. over $300 billion annually according to the FBI, yet receives less attention than blue-collar crime despite its massive societal impact. AI systems could potentially catch criminals like Bernie Madoff earlier by analyzing data more quickly than humans, detecting subtle patterns, and remaining immune to manipulation or distraction that might affect human investigators.
While we can never create a perfect justice system due to inherent human biases, AI can amplify our humanity and intelligence to create a fairer system. The goal is to fix the system's injustices in a cost-effective, attainable manner aligned with policy goals.
第 5 章
Journalism Reimagined: Truth-Seeking in the AI Era
American journalism has faced two decades of decline, with over 2,200 local papers closing and 40,000 newsroom jobs lost since 2005 as traditional business models collapsed under internet competition. Meanwhile, global information production has exploded to approximately 375 billion gigabytes daily. This information tsunami creates both challenges and opportunities for journalism institutions with their experience in storytelling, error-correction, and truth-seeking.
Speed has always been essential to journalism, described as "the first rough draft of history." Journalists constantly race against the clock to deliver timely information, making journalism not just a product but an iterative, self-correcting process. GPT-4 and similar AI tools will have an outsized positive impact by helping news organizations gather, produce, and distribute news faster-sifting through public records, monitoring social media, generating headlines, and transcribing interviews in seconds. While acknowledging GPT-4's tendency to hallucinate, journalism's established culture of verification makes it well-suited to manage these shortcomings, and the productivity gains justify the vigilance required.
GPT-4 provides a fundamentally different research experience compared to traditional web searching. Rather than offering links that may or may not be relevant, GPT-4 instantly synthesizes information from various sources, creating a starting point that orients the user quickly. This conversational approach creates a flow state where one question naturally leads to ten more, making the discovery process feel like steady progress rather than frustrating link-hopping.
While traditional news websites personalize content by tracking user behavior over time, a different approach using AI could emerge. Instead of surveillance-based personalization, news organizations could implement conversational interfaces where users actively shape their experience through queries like "Give me summaries of your three most-read tech stories" or "Create a counter-argument to today's op-ed using only your archive." This approach keeps journalists creating the foundational content while giving users more agency in consumption.
AI's potential for creating disinformation at scale remains a top concern, followed by job elimination. Steve Bannon's strategy to "flood the zone with shit" could be amplified by AI tools. While regulation may follow, the best defense is to "flood the zone with truth" through transparent, verifiable information systems. News organizations could adopt verification tools like "Fact Check" buttons powered by AI that would analyze articles' sources, statistics, and media elements, creating a clear distinction between content that welcomes verification and that which doesn't.
While AI creates opportunities for journalism, the industry's survival mode makes embracing change difficult. Journalists should experiment with AI tools while acknowledging that human judgment remains essential. Journalism demands curiosity, ethics, and empathy that AI lacks-GPT-4 can't investigate a fire scene or build source relationships.
第 6 章
Social Media Transformation: From Consumption to Creation
MrBeast (Jimmy Donaldson), a 24-year-old YouTube phenomenon with 221 million subscribers across his channels, exemplifies social media success with his talent for creating straightforward yet compelling video titles. When Reid experiments with GPT-4 to generate potential MrBeast video titles like "Lamborghini vs 10,000 Balloons: Can It Float?" and "I 3D Printed a Giant Ice Cream Cone and Ate It with a Forklift," he notes how creators are enthusiastically engaging with AI.
Social media has long incorporated AI through content moderation and recommendation algorithms, though these have created problems like filter bubbles and echo chambers. Until recently, users could only opt out of AI features rather than actively using AI tools like GPT-4 and DALL-E 2 in self-directed ways.
Social media's paradox is that while authenticity is its "coin of the realm," deception flourishes there too. Social media initially helped bring transparency online-LinkedIn connected real professional identities through networks of affiliation, while Facebook began with verified college emails. Yet as platforms scaled to hundreds of millions of users, governance became challenging. Facebook now removes over one billion fake accounts quarterly. Rather than pursuing hard-to-enforce bans on AI tools, we might consider "flooding the zone with truth" and developing new ways for creators to verify their humanity.
Reid envisions a future where LLM chatbots become a new media format for individuals to syndicate their thoughts, values and creativity across time and space-similar to books or podcasts. He illustrates this with "Codysseus," a fictional YouTube creator who could use his content archive to create a fine-tuned GPT-4 version of himself to interact with fans when he can't personally respond to all comments. Codysseus might create multiple specialized chatbots-one for coding instruction, another for digital nomad advice, even one for his cat-with some available only to paying subscribers or featuring sponsored content.
Just as YouTube evolved in unpredictable ways after its 2005 launch-becoming an educational resource, making cats into millionaires, and spawning entirely new content genres-AI will likely follow a similar path of creator-driven innovation. The chatbot examples represent just one possibility among countless creative approaches yet to be imagined.
第 7 章
Work Transformed: AI as Career Co-Pilot
The job landscape has evolved dramatically in recent decades-when Reid graduated college in 1990, roles like "web designer" and "data scientist" didn't exist, and when LinkedIn launched in 2003, there were no "social media managers" or "TikTok influencers." AI will drive similar transformations, rewarding companies and professionals who incorporate these tools most innovatively while leaving those who don't adapt struggling to remain competitive.
In his 2012 book "The Start-up of You" (co-authored with Ben Casnocha), Reid encouraged adapting to increased job mobility rather than clinging to outdated career models. AI can function as a career co-captain, serving as an on-demand advisor to help analyze situations and generate options when human advisors aren't available. While AI career advice may sometimes be generic, its potential value in career planning, skills assessment, job matching, resume optimization, and interview preparation is significant.
Reid's follow-up book "The Alliance" (with Ben Casnocha and Chris Yeh) reimagined the employer-employee relationship as an alliance between independent entities pursuing shared missions rather than as family (too permanent) or free agents (too transactional). AI could improve employee experience by offering feedback and development opportunities when needed, rather than being constrained by manager availability-especially valuable in busy modern workplaces with remote and hybrid arrangements.
In sales, while computers can't build rapport like humans, tools like GPT-4 can amplify human salespeople by handling time-consuming work many skip. AI can identify and qualify leads, analyze customer behavior, personalize pitches, track sentiment, and manage post-sale relationships. This allows human salespeople to focus on relationship-building and reading emotional cues. However, AI will likely reduce sales jobs, particularly entry-level Business Development Representatives, while increasing the productivity and compensation of those who remain.
In law practice, AI excels at analyzing large datasets, automating document and contract review, predicting case outcomes, and conducting technical analyses like patent searches. While AI might disrupt low-level legal work, it could actually improve job satisfaction by handling tedious tasks, allowing attorneys to focus on more complex and intellectually stimulating aspects of law.
In management consulting, AI could identify improvement opportunities, forecast market trends, generate insights from feedback, analyze competitors, and implement strategies. Early-career analysts face the greatest risk as their research and analysis tasks can be automated, while even partners might see reduced demand for conventional wisdom. Consultants who can't add value beyond standardized activities will suffer, while those using AI to generate unconventional insights will thrive.
第 8 章
The Hallucination Challenge: When AI Makes Things Up
When ChatGPT launched in November 2022, OpenAI warned it might produce "plausible-sounding but incorrect" answers. These concerns about AI hallucinations echo similar criticisms once leveled at Wikipedia and the early internet. Reid has experienced GPT-4's fabrications firsthand-inventing podcast details, generating fake citations, and confidently asserting contradictory information.
The term "hallucinations" covers at least four distinct types: nonsensical outputs (easily identified), plausible but incorrect information (most problematic due to AI's convincing authority), claims of capacities the AI doesn't possess, and deliberately generated false information prompted by users with malicious intent. These hallucinations attract attention partly because they contradict our expectations of how advanced AI should behave.
Despite being inundated with complex and sometimes misleading information daily, access to "settled truths" has enormous value. Wikipedia exemplifies this-once criticized as unreliable, it now receives billions of monthly pageviews. As founder Jimmy Wales noted, it provides "good enough knowledge, depending on what your purpose is." Wikipedia succeeded through superior distribution-being free, accessible, comprehensive, and constantly updatable-which outweighed initial quality concerns.
Reid sees LLMs following a similar path as Wikipedia-despite occasional errors, they provide "good enough knowledge" with significant advantages in breadth, efficiency, and accessibility. While some call for immediate regulation, we need more real-world usage data before determining appropriate rules. The digital nature of LLMs allows for rapid fixes when problems arise, as demonstrated by Microsoft's quick response to Bing Chat issues.
There's an important distinction between problematic hallucinations and productive imagination. When GPT-4 creates fictional stories or speculative business plans, these "made-up" texts amplify our human capacity for creative thinking. What we ultimately want is an LLM that functions factually in some contexts and imaginatively in others, while clearly indicating which mode it's operating in. Despite current caution needed with GPT-4, future versions will likely significantly decrease unintended hallucinations while preserving creative capabilities.
第 9 章
Homo Techne: Technology as Our Defining Feature
Reid discusses how technology makes us more human, referencing a 1920s technological optimism that flourished despite recent tragedies like World War I and the Spanish Flu pandemic. Americans embraced innovation, quadrupling electricity production and developing various networks. He identifies himself as a "techno-humanist," arguing that technological innovation is humanity's defining feature, suggesting we might more accurately be called "Homo techne" rather than "Homo sapiens."
Despite technological pessimism, humanity has accomplished remarkable things through technology. Over thirty years, global poverty rates have plummeted, child mortality declined by half, and air pollution in the US dropped 78% since 1970. RNA therapeutic advances enabled rapid COVID-19 vaccine development. However, this progress creates an irony: we become so accustomed to technological marvels that anything less feels like failure.
Reid expands on his concept of "Homo techne" by discussing Lucy, a 3.2-million-year-old Australopithecus afarensis fossil that rewrote our species' self-definition by elevating tool use over brainpower. He argues that tool use stimulated our cognitive evolution, with tools allowing more efficient food gathering, leading to dietary changes promoting brain growth while stimulating neurological development.
Reid challenges the view of technology as dehumanizing, arguing instead that it's what makes us fundamentally human. He questions the artificial distinction between arts and sciences, noting that all human expression-from art to communication to career work-is enabled and extended by technology. Rather than fearing that AI and other emerging technologies will diminish humanity, Reid suggests that future humans will be more human, not less, because of technological evolution.
Technologies are never neutral-they embed specific values and intents-but this doesn't make them inherently good or bad. Reid's techno-humanist approach involves striving for outcomes that broadly benefit humanity while remaining vigilant about potential negative effects. He warns against both evading responsibility by claiming technology is neutral and invalidating technologies simply because they might produce negative outcomes.
Reid argues against preemptive AI bans that lock in current inequities and unrealistic error-free standards that prevent progress. Instead, he advocates for accelerating technological innovation to address global challenges like climate change, resource depletion, and institutional distrust. While acknowledging his perspective might seem optimistic, Reid believes embracing our innovative nature with bold ambition represents humanity's best path forward.
第 10 章
At the Crossroads: Choosing Our AI Future
Our current moment represents a pivotal crossroads where humanity must choose how to harness AI's potential. While acknowledging that both great achievements and significant problems will inevitably come with AI, we must focus on designing AI systems that broadly empower all humans rather than just a privileged few.
Reid argues that embracing AI requires accepting both inevitable failures and necessary regulation. Drawing parallels to fire, wheels, and even bathtubs-all revolutionary technologies that brought both benefits and dangers-he emphasizes that zero risk is only possible with zero progress. While we're still in the "early, early days" of AI development, patience and perseverance are essential. Just as we accept certain risks with smartphones because of their immense utility, society will need to determine what level of risk is acceptable with AI technologies.
Reid warns against a potential dystopian future where humans become passive consumers completely dependent on AI-a world where people use AI for everything from writing marketing memos to making major life decisions like when to have children, leading to "completely de-risked lives of statistically probable stability." When Reid asks GPT-4 to describe this future in Thoreau's style, the AI responds with a warning about machines reducing humans to "passive consumers and spectators" who have forgotten "what it means to be human."
As AI improves, we must "demand more from ourselves" and position these tools as collaborative partners rather than replacements. The choice is clear: use AI to help us do less, or use AI to help us do better. Reid remains optimistic that humanity will choose the path of progress, continuing our evolution as Homo techne-creatures defined not just by our wisdom but by our unique ability to create and use tools that expand our capabilities beyond what nature alone provided.
