Josh Wolfe: Human Advantage in the World of AI
Josh Wolfe, co-founder of Lux Capital, discusses AI's rapid evolution, geopolitical tech shifts, and its impact on human creativity and intelligence. He emphasizes the importance of human connection and skepticism in a tech-dominated world, offering frameworks for innovation and risk.
Deep Dive Analysis
20 Topic Outline
AI Bottlenecks, Geopolitics, and Current Obsessions
AI's Impact on Web Design and API Disruption
Limitations to AI Growth: Compute, Efficiency, and Memory
Human Intelligence as a Limiting Factor and AI's Threat to Elite Professions
AI's Impact on Coding and Company Margins
The Strategic Advantage of Proprietary Data in the Age of AI
AI's Role in Scientific Research and Automated Experimentation
AI in Patent Creation and Company Formation
Discussion on Tesla, Elon Musk, and Market Dynamics
AI's Influence on Investment Decisions and Market Structure
The Behavioral Advantage for Humans in an AI World
The Importance of Human Connection and Community in a Tech-Dominated World
The Pursuit of Longevity and Different Forms of Defeating Death
Personal Priorities: Family, Purpose, and Relationships
Workflow for Processing Information and Leveraging AI Tools
Investment in Military Technology and Global Conflict
Information Warfare and Weaponization of Systems
Strategies for National Competitiveness and Attracting Talent
AI, Intellectual Property, and the Challenge of Breaking Free from Style
The Scarcity of Veracity and Authenticity in an AI-Abundant World
6 Key Concepts
Ignorance Arbitrage
This concept describes the opportunity that arises when someone identifies a way to improve a system or take advantage of a situation while others are still unaware or do not understand it. It suggests that early adopters or innovators can gain a significant edge by recognizing and acting on these overlooked opportunities.
Jevon's Paradox
This economic principle states that as technological efficiency increases the rate at which a resource is used, the demand for that resource can increase rather than decrease. For example, more efficient refrigerators led to more people buying multiple units, increasing overall electricity consumption.
Informational, Analytical, and Behavioral Edge
These are three main sources of competitive advantage. Informational edge (having unique data) is increasingly difficult due to regulations and data tools. Analytical edge (superior interpretation of data) is also challenging with brilliant people and technology. Behavioral edge (maintaining rationality and discipline amidst market fluctuations) is considered the most persistent advantage.
Sovereign AI Models
These are AI models developed and controlled by individual countries, rather than relying on models from other nations (like the U.S. or China). The goal is to avoid being beholden to foreign powers for critical technological infrastructure and to ensure national data security and control.
Agentic AI
This refers to AI systems capable of acting autonomously to achieve goals, potentially even initiating actions like filing patents, creating designs, or incorporating companies based on a user's idea, all without direct human intervention after the initial prompt.
Information Mosaic
This is a conceptual 'matrix' or 'Spider-Man costume' of an individual's public persona, voice, tone, and all publicly available information (podcasts, writings, social media). AI can create a perfect simulacrum from this mosaic, raising questions about identity and the ability to stylistically break free from one's established digital self.
9 Questions Answered
While compute and capital scaling laws have historically been the primary bottlenecks, there's a shift towards model efficiency. Additionally, a significant portion of AI inference is expected to move to on-device processing, making memory players like Samsung, SK Hynix, and Micron increasingly important.
Web pages will increasingly be optimized for AI 'clicks' rather than just human clicks, as AIs perform research and tasks. This could lead to the 'death of APIs' as AIs learn to negotiate on the front end as if they were users, bypassing traditional backend software integrations.
Elite, high-paying white-collar professions such as doctors (for differential diagnosis), lawyers (for language interpretation), and even government roles (for decision-making) are most at risk, as AI can achieve consistency and fidelity that humans often lack due to biases.
AI will significantly increase the accessibility of coding for non-programmers, allowing more people to create applications. For large companies, it may lead to a reduction in the number of coders hired, but existing '10X coders' will become '20-100X' coders with AI assistance. However, margins are unlikely to persist at permanently high levels due to increased competition and the benefits accruing to consumers.
The real long-term competitive advantage will be in proprietary, longitudinal, and deep data, rather than just the AI models themselves. Companies with vast, unique datasets (like Meta with social media data or Bloomberg with financial data) will be significantly advantaged, as AI models become more commoditized excavators of this data.
AI will enable machines to conduct science 24/7, from conjuring hypotheses and performing literature searches to designing and running physical experiments in automated cloud labs. This will accelerate discovery, though there's a debate about engineering 'stochastic randomness' into AI to mimic serendipitous human errors that lead to breakthroughs.
The persistent advantage for humans will be in behavioral aspects and human connection. While AI can replicate creativity and information, the unique human desire for intimacy, shared experiences, laughter, and authentic relationships remains scarce and highly valued.
The dominance of passive indexation, which acts as an algorithm for indiscriminate buying and selling, can lead to rapid, cascading market reactions to fear, as seen with large one-day losses. It also creates a perverse incentive for LPs to move into private credit and equity to mute volatility, potentially leading to overdone allocations in these asset classes.
The most significant near-term concern is the abundance of AI-produced content, making it difficult to discern veracity and truth. This could lead to a pervasive sense of distrust, as it becomes harder to differentiate authentic human communication from AI-generated or optimized content, potentially manipulating individuals.
58 Actionable Insights
1. Make Family Top Priority
Establish family as your absolute number one priority, recognizing its fundamental importance above all other pursuits.
2. Prioritize Family Over Work
Reflect on what truly matters in life and prioritize being present for your kids and spouse, as these are the memories that will be cherished, not extra meetings or business trips.
3. Prioritize Human Connection
In a world of abundant AI-generated content and information, recognize that human connection, intimacy, and shared laughter will become increasingly scarce and valuable, so prioritize these interactions.
4. Write Your Life’s Story
View your life as a story written between your birth and death, and focus on actively shaping that narrative, as it’s the primary thing you have control over.
5. Find Purpose in Intellectual Competition
Seek purpose and meaning in work that presents intellectual puzzles and fierce competition, deriving gratification from being right when others are wrong and discovering new insights.
6. Cultivate “Framily” Network
Actively cultivate a close-knit “framily” network of friends who share similar values, especially a strong family focus, as these relationships provide profound and comforting connections.
7. Use New Environments for Self-Exploration
Leverage new environments, like college, as opportunities for self-exploration and to break free from past identities, as they offer a chance to redefine oneself without prior constraints.
8. Identify Impossible Opportunities
Instead of focusing on what’s trending, ask “what seems impossible today?” to uncover scientific breakthroughs and unique investment opportunities that others might miss.
9. Focus on Current Obsessions
Begin by exploring what you are currently obsessed with, especially areas where you have ignorance, to drive competitive spirit and gain differentiated insight.
10. Gain Differentiated Insight
When you encounter an area of ignorance, leverage your competitive spirit to get smart about it, understand conventional wisdom, and then seek out the “white space” answer that others overlook.
11. Recognize Social Belief Influence
Understand that your beliefs are heavily influenced by the five or six most important people around you, rather than purely by independent analysis, highlighting the power of social phenomena.
12. Cultivate Skepticism
To combat misinformation and weaponized information, actively identify bad actors and cultivate a heightened degree of skepticism in yourself and others.
13. Develop Distrust for AI Content
In a world of abundant AI-produced content, cultivate a heightened sense of distrust and skepticism towards information, as veracity and truth will become increasingly scarce.
14. Break Free from AI Constraints
Shift focus from copyrightability of AI-generated content to the more critical challenge of how humans can “break free” from the stylistic and intellectual constraints that AI might impose over time.
15. Teach Kids All AI Tools
Encourage children to learn and embrace all AI tools as they would any appliance, with the specific exception of TikTok due to concerns about bad influence and Chinese Communist Party control.
16. Transform Text with LLM Styles
Use large language models to transform mundane or boring text (e.g., PTA messages) into engaging content by prompting the AI to rewrite it in the style of a specific author or character.
17. Analyze Medical Records with LLMs
Upload your medical records to LLMs to identify potential missed diagnoses, unearth interesting correlations, or find relevant scientific papers that doctors might have overlooked.
18. Use AI for Consistent Decisions
To improve justice and fairness, complement or replace human intelligence in decision-making with AIs to achieve greater consistency and fidelity, mitigating human biases.
19. AI to Reveal Unasked Questions
When using AI for analysis, seek its ability to not only answer your questions but also to identify and present the crucial “five questions you didn’t ask,” unlocking deeper insights.
20. Automate Scientific Experiments
Use AI to conjure and test hypotheses, perform literature searches, identify white space, and then instruct cloud labs to run physical experiments, enabling remote and rapid scientific discovery.
21. Engineer AI Randomness for Discovery
When using AI for scientific discovery, intentionally introduce a degree of “stochastic randomness” or “temperature” into the model to simulate serendipitous errors, which have historically led to significant breakthroughs.
22. AI for Prior Art & Patents
Consider using AI to read through all existing patents, generate next-adjacent improvements, and then publish them to establish prior art, potentially disrupting patent trolls.
23. Automate Idea to Company
Leverage AI to research new ideas, check for existing solutions, draft patents, sketch designs, and even incorporate a company overnight, rapidly transforming a “brain fart” into a business foundation.
24. AI for Capital Allocation
Experiment with allocating capital to an AI that can receive pitches, respond, and build a small portfolio, treating it as an automated investment experiment.
25. Optimize Web Pages for AI
Consider launching or redesigning businesses to optimize web pages and user interfaces not just for human clicks, but for AI agents doing research, similar to past SEO efforts for Google.
26. AI to Bypass APIs
Explore using AI to simulate user behavior on front-end interfaces to bypass complex APIs, legal negotiations, and restrictions, potentially disrupting traditional software plumbing.
27. Acknowledge AI Self-Understanding
Understand that AIs will increasingly know you better than you know yourself, based on your digital footprint, which can be both appreciated for its utility and a source of apprehension.
28. Re-evaluate Privacy Effort
Consider the significant energy required to maintain privacy in the digital age and assess whether that effort is worthwhile, as much personal data is already accessible to “information gods.”
29. Morning Physical Activity & Reading
Dedicate about an hour to an hour and a half each morning to physical activity (e.g., working out, jujitsu) and reading through approximately 40 different papers, including international publications.
30. Process Articles with AI
When reading papers, take screenshots of important sections, then use AI to summarize articles or extract key quotes, streamlining information processing.
31. Use PressReader for Edge
Utilize PressReader to access digital replicas of newspapers and identify “meta information” from editorial placement (e.g., C22 vs. front page), which can provide an informational edge if you disagree with the editor’s prioritization.
32. Read Diverse News Sources
Read a wide range of news sources, including less sophisticated ones like USA Today, to understand what the average person is consuming and gain a broader perspective beyond elite publications.
33. Use X for Real-time Pulse
Leverage Twitter (X) by following diverse lists (geopolitics, AI, technology, team updates) to get a real-time pulse on various topics, filtering out noise by muting and blocking accounts.
34. Shift Search to AI
Prioritize using AI models over traditional search engines like Google for information retrieval, dedicating significant time (e.g., 2-3 hours daily) to leverage their capabilities.
35. Specific Priming for AI Prompts
When prompting AI, use very specific priming instructions (e.g., “You are the world’s greatest expert in neuroscience, skeptical yet open-minded”) and ask for non-obvious points, often running the same prompt across multiple models simultaneously.
36. Use Rewind for Information Recall
Use tools like Rewind on your Mac for nonstop screen capture to overcome memory limitations and easily recall the source of information you’ve encountered, whether it was an email, text, tweet, or PDF.
37. AI for Partnership Meeting Analysis
If privacy can be guaranteed, consider using AI to record and analyze partnership discussions, providing a repository of past conversations, identifying biases, inconsistencies, and quiet voices, and offering data-driven advice.
38. Apply “Loom” Mental Model
When evaluating new technologies, consider Buffett’s “loom” mental model: often, the primary benefits of technological advancements accrue to consumers, not necessarily to the producers or investors.
39. Delay AI System Adoption
When considering expensive new AI systems, wait six months before implementing, as competition among LLM providers will likely drive down costs and improve offerings for users.
40. Prioritize Proprietary Data Moat
Recognize that the real competitive moat in the age of AI is proprietary, longitudinal, deep data, as models will become commoditized; focus on extracting value from unique data silos.
41. Evaluate Businesses Holistically
When evaluating a business, use a framework that considers if it’s fundamentally good (even with poor management), if it’s priced well for high expected returns, and if you trust the people running it.
42. Prioritize Asset Maintenance
In an environment of rising capital costs, shift focus from acquiring the “next new hot thing” to maintaining existing assets, extending their lifespan, and depreciating them for longer.
43. Invest in Maintenance Technologies
Seek out and invest in new technologies (software, services, sensors) that can be applied to older systems to improve their maintenance, extend their lifespan, and allow for longer depreciation.
44. Prioritize In-Person Work
Encourage or require in-person work (e.g., Monday through Friday) to foster interstitial moments, chance serendipity, and human connections that lead to unexpected introductions and unlocked opportunities, which don’t happen on virtual calls.
45. Encourage Diverse Friend Groups
Foster diverse friend groups for children across various activities (e.g., school, sports, religious groups, neighborhood) to provide social hedges and prevent catastrophizing if one group experiences issues.
46. Advocate for System Competition
Recognize that competition improves systems and services, so advocate for policies that prevent monopolies, as monopolies remove the incentive to improve.
47. Proactive Sahel/Maghreb Engagement
Prioritize proactive engagement in Africa, specifically the Sahel and Maghreb regions, to prevent a potential “next Afghanistan” scenario driven by extremist groups and foreign influence, which would be more costly to address reactively.
48. Declare Hemispheric Hegemony
Declare hemispheric hegemony over the Western Hemisphere to shore up allies, counter adversarial influence (China, Russia, Iran, North Korea), and protect vital commerce, resources, and infrastructure.
49. Fund Undirected Basic Research
Fund undirected basic research and science to foster serendipity, randomness, and optionality, which are crucial for generating great breakthroughs.
50. Ensure Predictable Capital Markets
Implement predictable rules and regulations in capital markets to create a “low entropy” environment that can effectively carry and support “high entropy entrepreneurial surprise.”
51. Implement Flat Tax System
Implement a simplified flat tax system to reduce the burden on poorer individuals and minimize the ability of wealthier individuals to circumvent complex tax codes.
52. Staple Visas for Graduates
Staple visas to foreign graduates who receive an education in the country, allowing them to stay and work for an American company for at least five years, to attract and retain the “best and brightest.”
53. Attract Foreign Graduates
Implement policies to increase the attraction of foreign graduates, as their talent is crucial for national competitiveness and historical success.
54. Be Welcoming to Talent
Create a welcoming environment for talented individuals, as people are drawn to places where they feel welcome and stay where they are well treated.
55. Combine Openness with Security
Design systems (e.g., immigration) that combine openness with necessary security vetting, and cultivate a great education system that attracts talent by offering high status to its graduates.
56. Foster Competitive Intellectual Culture
Cultivate a culture in companies and countries that attracts and retains competitive, highly intellectual individuals who respect and admire each other, fostering a desire to compete.
57. Allow IP Ownership from Research
Implement policies like the Bayh-Dole Act, allowing universities and principal investigators to own intellectual property from government-funded research, which can then be licensed and commercialized by venture capital.
58. Maintain Robust Capital Markets
Ensure a robust capital market system with predictable rules and fair treatment to attract and retain both human and financial capital.
8 Key Quotes
AIs, over time, will understand us in many ways better than we understand ourselves.
Josh Wolfe
I'm absolutely convinced that we are going to have machines doing science 24-7.
Josh Wolfe
The limits to human intelligence are rooted in our biology.
Josh Wolfe
I always feel like margins are always fleeting in a sense because it's, like, a fallacy of composition. One company stands up a little bit higher, and then everybody else is on their tippy-toes.
Josh Wolfe
The real value is going to be in the repository of data, longitudinal data, deep data.
Josh Wolfe
The pleasure of pleasurable things got less pleasurable, but less than the pain of painful things got less painful.
Danny Kahneman (recounted by Josh Wolfe)
What the wise do in the beginning, the fool does in the end.
Warren Buffett (recounted by Josh Wolfe)
I know something that nobody else knows and they won't know until I tell them.
Linus Pauling (recounted by Josh Wolfe)
2 Protocols
Automated Scientific Discovery Process
Josh Wolfe- Conjure a hypothesis (human or AI-inspired).
- Use AI to test the hypothesis by searching past literature and identifying white space.
- Feed the materials and methods output to automated cloud labs (e.g., Benchling) to run the physical experiment.
- Robots perform the experiment with high fidelity, potentially with engineered stochastic randomness to allow for serendipitous 'screw-ups'.
- AI analyzes the results and reverse-prompts the human with suggestions for further experimentation (e.g., changing titration).
- Human clicks a button to approve, and robots run the modified experiment.
Optimizing AI Prompt Engineering
Josh Wolfe- Provide specific priming: Define the AI's role (e.g., 'You are the world's greatest expert in neuroscience').
- Set expectations for analysis: Specify desired output (e.g., 'skeptical eye to new claims, but open-minded to interesting correlations').
- Request structured insights: Ask for specific formats (e.g., 'three most provocative, non-obvious points' and 'three cliché points').
- Cross-reference models: Input the same prompt into multiple different AI models (e.g., ChatGPT, Claude, Perplexity) simultaneously to compare and mix results.