Summary

Nate Silver explores "The River" – an ecosystem of risk-takers unified by probabilistic thinking, EV, and game theory. Through examining the 13 habits of highly successful risk takers, Silver shows how they combine being cool under pressure with strategic empathy, process-oriented thinking, and conscientious contrarianism. He critiques EA and rationalism for trying to quantify the unquantifiable and ignoring Chesterton's fence, while exploring concepts like the Kelly criterion for managing risk and the orthogonality thesis in AI safety. The book concludes with three principles to compete and prosper: agency (autonomy and optionality), plurality (preventing ideological dominance), and reciprocity (treat others as intelligent, play the long game).

"We often spend the most time debating the least important decisions"

"It's easier to achieve alpha at your objective when others aren't playing the same game"

"Look for flaws in incentives, not intelligence"

"The richest founders in the world are… highly skilled degenerate gamblers who got lucky"

Key Takeaways

• Game theory: max EV given other agents. Nash equilibrium. Randomization is essential to GTO
• GTO vs exploitative strategy - exploit and you risk being exploited
• Poker involves two skills rarely found together: systems thinking + empathy
• Better to view yourself as lucky: constantly try new things, listen to intuition, have positive expectations, resilience

• Keynesian beauty contest: picking most popular rather than best option → group think, conformity
• VC as unique: very long time horizons, asymmetric odds
• Founders = hedgehogs, investors = foxes. Foxes need to thrive on imperfect info

• Kelly criterion: maximizing long-term EV while minimizing risk of ruin (edge / odds)
• Orthogonality thesis: an AI's intelligence and goals are uncorrelated
• Modern AI models don't need domain knowledge → low interpretability

Notes

0: Introduction

• People are becoming more bifurcated in their risk tolerances
• Domain-specific risk taking
• Expected value (thinking probabilistically)
• The River: ecosystem of people and ideas
  • EV, Bayesian priors
• Upper river: rationalism, EA
  • NorCal, AI
• Mid river: VC, hedge funds
• Down river: casino gambling
• Archipelago: online sports betting, crypto
• 2 clusters of attributes for success:

1. Cognitive: abstract analytical reasoning
   • Root cause analysis
   • Model building, heuristics
   • Decoupling: playing devil’s advocate
     • “Yes, but” statements
2. Personality: competitive, independent
   • Contrarian: dislike consensus thinking
   • Risk tolerance
   • Free speech advocates

• The Village: gov’t, media, mostly dems
• Critique of the village:
  • Too political
  • Couplers
  • Conformist
• Critique of the river:
  • Unregulated capitalism
  • Rugged individualism

1:  Optimization

• Tight, aggressive poker
• 80s were primordial soup of poker
• Early 2000s solver revolution
• GTO vs exploitative strategy
• ↑ eg. rock paper scissors:
  • GTO: 33%, exploitative: 45%
• Game theory: max EV given other agents
• Nash equilibrium
• Randomization is essential to GTO
• ↑ “We often spend the most time debating the least important decisions”
• Exploit and you risk being exploited

2: Perception

• Whales + professionals dynamic
• Most people don’t take enough risk
• Most successful traders:
  • Higher testosterone
  • Higher in-the-zone risk response
• “There’s nothing like paying to build up your pain tolerance”
• Poker involves two skills rarely found together: systems thinking + empathy
• Gut instinct
• System 1 + 2 thinking
• Top 200 poker player won’t break even:
  • Over 1 year 50% of the time
  • Over 10 years 11%
  • Over 50 years 0.5%
• Better to view yourself as lucky:
  • Constantly try out new things
  • Listen to intuition
  • Have positive expectations
  • Resilience
• Flow: more easily achieved when facing intense risk?

3: Consumption

• Gambling is a privilege, not a right. Thus, casinos control who gets to play
• Rise of the river underdogs, eg. Nate Silver predicting every state right in 2012 election
• History of Las Vegas*
• Market for lemons: prisoners dilemma of taking most risk to profit off bad product
• Business of gambling*
• Trump is neither river nor village
• Rewards programs*
• Economist’s fable: “finding $20 bills that aren’t supposed to exist”
• Lottery + horse gambling = tax on poor
• Flow of slots: desire to escape, not win

4: Competition

• Top-down vs bottom-up sports betting
• Betting lines are still mostly manual due to things like player injuries, manipulation
• Large attack surface
• Replication crisis:
• Super Bowl is gold for sports casinos
  • Best bettors tip their hand at start
  • Casual bettors bet on optimal lines
• Market maker vs retail books
• Good book-making: get to the closing line as fast as possible, using professional bettors as consultants
• 3 skills needed:
  • Market skills (eg. trader)
  • Analytical skills (eg. stats)
  • Sports domain knowledge
  • Also networking, deal flow
• Providing books value vs steam-chasing
• Whale “bearding”: messenger betting
• Sports betting is an info hunt

13: 13 habits of highly successful risk takers

1. Cool under pressure
   • Don’t try to be a hero
2. Courage, competitive, bring it on attitude
3. Strategic empathy
4. Process-oriented, not results
   • Self-awareness, humility
5. Take shots, okay with failure
6. No mediocrity, half way: raise or fold
7. Prepared, well-trained
8. Selectively high attention to detail
   • Conserve attention, focus, bandwidth
   • What’s the main thing?
9. Adaptable: good generalist
10. Good estimators: Bayesian
    • Also quick to estimate, act
11. Try to stand out: independent mind, purpose
    • Don’t want to jump through society’s hoops
12. Conscientiously contrarian
    • Contrarian vs independent
    • Have theories about why the market is wrong
    • “It’s easier to achieve alpha at your objective when others aren’t playing the same game”
    • “Look for flaws in incentives, not intelligence”
13. Not driven by money. Live on the edge because that’s their way of life

5: Acceleration

• Silver worries about embrace of “accelerationism”: careful with AI
• Silicon Valley has to be full of both rationalists and unreasonable-s (founders)
• History of Silicon Valley*
• Creative destruction for the sake of creative destruction
• Why VC is unique:
  • Very long time horizons
  • Asymmetric odds
• Moral hazard (eg. bank bailouts)
• Lack of fear of looking stupid, loss aversion
• Founders = hedgehogs, investors = foxes
• Foxes need to thrive on imperfect info
• Democrats: skepticism → dogmatism
• Silicon Valley as an idea lab (Tim Urban)
• More self-made billionaires than ever before (70%), though lower overall social mobility
• Optimal amount of adversity
• “The richest founders in the world are… highly skilled degenerate gamblers who got lucky”
• Keynesian beauty contest: picking most popular rather than best option
  • Leads to group think, conformity
• VC prominence is sticker than other industries (brand importance)
• a16z / “top decile” portfolio returns:
  • 25% 0x
  • 25% 0-1x
  • 25% 1-3x
  • 15% 3-10x
  • 10% 10x+
• ↑ simulation:
  • 96% of firms beat inflation
  • 90% beat S&P500
  • Average IRR: 24%
• Simulation protagonist theory:

6: Illusion

• SBF is an outlier in the river
• Multiple personalities
• Crypto culture in Miami*
• Meme creation of value:
  • eg. shit-coins, #WSB
• Options, margins trading
• “If you have a gambling problem, then someone is going to come up with some product that touches your probabilistic funny bones.”
• Vitalik’s math: Bitcoin as 10% of gold
• Focal point: in game theory, a choice we can practically coordinate
  • Eg. NYC Grand Central Station
  • Eg. Bitcoin gold standard
• Memetic desire: ↑, envy-based economy
• ↑ for SBF*

7: Quantification

• EA and “earning to give”*
• Rationalism:
• Great man theory: Elon and Bezos
• Trolly problem*
  • Eg. dog lost on NYC train tracks
• Argument against EA: can’t quantify the unquantifiable
  • Eg. correlating value with neuron count or intelligence is a slippery slope
• Covid and cost-benefit analysis*
• Value of statistical life (VSL): $10m in USA
  • Based on revealed preferences
• EA: “a movement that tries to figure out, of all the different uses of our resources, which uses will do the most good, impartially considered”
• Rationalism: applying scientific thought to everything
• Impartiality:
• Utilitarianism: “the greatest amount of good for the greatest number”
• Infinite ethics:
• Critiques of utilitarianism: ↓
• Almost all models under-simplify
  • eg. dogs vs pigs neuron count
• Hedonistic, preferential utilitarianism:
• Study: philosophers are as rational, objective as scientists
• Solution: could assign weights to different moral frameworks
• “Sometimes, EA seems like a moral framework for a world where everyone has 200 IQs and not the world we actually live in”
• The fallacy of Chesterton’s fence: know why rules are in place before removing
  • Reciprocity: can’t assume everyone is following the same moral code
  • Rule utilitarianism: ↑
• P(doom), Manifold platform*
• Instrumental rationality: acting in consistency with your beliefs and goals
  • Consistent preferences:
• Epistemic rationality: your actions and beliefs line up with reality
• NPC syndrome:
• Futurism + Transhumanism:
• Futarchy: prediction markets → policy

8: Miscalculation

• SBF’s excuses*
• 4 theories of SBF: *
• Kelly criterion: maximizing long-term EV while minimizing risk of ruin
  • Basically edge / odds
• Simulation using % of ↑ for NFL betting:
  • 1x Kelly: ahead 80% of time
  • 5x Kelly: ahead 10% of time
  • ↑ Higher EV if don’t care about ruin

♾️**: termination**

• Sam Altman and OpenAI*
• Roon, Beff Jezos, e/acc*
• Orthogonality thesis: an AI’s intelligence and goals are uncorrelated
• Von Neumann, MAD*
• “We foxes think pessimism bias and optimism bias are mistakes and equal measure”
• MAD shouldn’t work because a counterstrike lowers EV to near -♾️
  • Though it does due to type I thinking
• Prospect theory: ↑
• Modern AI models don’t need domain knowledge → low interpretability
• Bid-ask spread:
• Tetlock’s p(doom) debate results:
  • Domain experts: 8.8%
  • Generalists: 0.7%
• Reference classes: ↑ , eg. “just math” vs “new species of life”
• Technological Richter scale:
  • 4 = patent, valuable IP
  • 5 = commercially successful invention
  • 6 = shortlist for tech of the year
  • 7 = one of decade’s leading inventions
  • 8 = eg. auto, electricity, internet
  • 10 = defines a new epoch (singularity?)
• ↑ experts put AI anywhere from 7-10
• Reasons for high p(doom):
  • Competition
  • Low interpretability
  • Public institutions dysfunctional
  • No unified human values framework
  • Should trust domain experts more
• Reasons for low p(doom):
  • Tech underestimates imminent popular AI backlash
  • Experts underestimate the scope of intelligence (language vs robotics)
  • Scientific, economic progress faces a lot of headwinds
• Secular stagnation: ↑
• “Civilization needs to learn to live with the technology we’ve built, even if that means committing ourselves to a better set of values and institutions.”

1776: Foundation

• 1776 as start of Industrial Revolution*
• Progress is not a given
• 3 principles to compete and prosper:
  • Agency: having good options
    • Autonomy, optionality
  • Plurality: not letting any one group or ideology gain a dominant share
    • Moral parliament
  • Reciprocity: treat other people as intelligent, capable
    • Play the long game (don’t exploit)