AI summary by TwinMind
Manifold Markets Founder - Predictions Markets & Revolutionizing Governance
Dwarkesh Patel
50m 53s
Introduction and Background
- Steven Gruget, co-founder of Manifold Markets (one of three co-founders)
- Manifold received grant from Scott Alexander, raised $2M seed round
- Building platform for user-created prediction markets
- Anyone can create questions about anything, others bet on outcomes
- Market mechanism aggregates information to produce calibrated probabilities
Play Money Currency System
- Uses play money currency (Manifold dollars) instead of real money
- New users receive 1,000 Manifold dollars upon signup
- Hypothesis: people driven more by status/competitiveness than greed
- Platform functions as social game for prediction skill development
- Users demonstrate knowledge through objective track record from betting history
Top Predictors and Talent Identification
- Current top predictor won big on Russia-Ukraine invasion market in February
- Overlap between forecasting community and financial professionals
- Experience is key factor for top predictors
- First hurdle: getting used to clarifying/crystallizing views about world
- Must actively put out numerical predictions and record them for improvement
- Untapped potential in top predictors that sites like Manifold can identify
Technical Infrastructure and User Experience
- Originally started as crypto idea before pivoting to web 2.0
- Crypto concerns: onboarding friction, MetaMask wallet requirements, transaction confirmation delays, transaction costs
- Only tiny chunk of global population has crypto wallets
- Current process allows betting within 30 seconds of visiting site
- Real money crypto offering on back burner as separate product
Market Creation and Resolution
- Market creators resolve questions themselves (user-resolved markets)
- Bimodal distribution expected: few huge high-trust creators with high volumes, long tail of smaller individuals
- Smaller creators' markets participated in by people closer on social graph
- Key insight: user-resolved markets can work despite fraud potential
- Small amount of fraud but manageable if users choose which markets to participate in
Corporate Prediction Markets
- Many companies have tried internal prediction markets (Google, GM, CIA)
- Most abandon them despite praising benefits
- Main reason: managers don't want negative feedback about their chosen course/vision
- Even positive feedback can cast doubt and lower mission success probability
- Rational concern: prediction markets can undercut management mission
- Better use cases: informational aspects, market research, consumer behavior prediction
- Usability concerns: hard to create simple, easy-to-use product for all employees
Decision Making and Information Aggregation
- Difficult to operationalize success criteria into single measurable thing
- Prediction markets work best when people create markets on variety of metrics
- Human intuition still needed to determine best course of action using market information
- Markets excel at isolating particular risks you care about
- Multiple well-targeted proxy variables better than broad financial instruments
Long-term Vision and Future Applications
- 10-20 year vision: embedded prediction markets in news articles, blog posts, CNN
- Allow people to immediately get sense of grounded opinions
- Make conversations more productive and fact-based
- Pessimistic take: people don't consume politics to understand what's happening
- Small portion of news consumption is for legitimate world understanding
Platform Mechanics and Fees
- 4% of trades go to market creator, 1% burned
- Fees impede market efficiency but incentivize desired behaviors
- Creator fee encourages market creation and resolution
- Liquidity fee feeds into liquidity pool, subsidizing entire market
- Tournaments with real cash prizes planned despite play money product
Addressing Long-term Questions
- Fundamental human constraint: care less about far future events
- Interest rate/discounting is facet of human nature
- Can't bet on apocalypse scenarios (won't be there to collect winnings)
- Brian Caplan/Eliezer Yudkowsky bet structure: pay now, collect later if no catastrophe
- Break down long-term questions into shorter-term proxy variables
- Example: Instead of "AI apocalypse 2050," ask about GPT-4 benchmarks, polling numbers
Market Efficiency and Whale Problem
- Pro efficient markets perspective
- Whales betting wrong side provide subsidy to correct predictors
- Markets efficiently solve whale problem by redistributing capital
- Direct analogy to real markets with pension fund rebalancing
- Not interested in quadratic schemes that would inhibit price discovery
Incentives and Virtual Economies
- People willing to invest heavily in virtual economies (World of Warcraft, Second Life)
- Equivalent of hundreds of thousands/millions of dollars in game worlds
- Human ability to focus on interests very powerful even without financial payout
- Global leaderboard becomes less important as site grows
- Personal markets and friend networks more relevant than global rankings
Company Operations and Internal Use
- Actively create markets on company-relevant questions
- Market on completing fundraising round by end of April (stayed above 85%)
- Market on onboarding 3 employees before end of June
- Early monetization decision informed by conditional prediction market
- Dog-fooding approach to using own platform
Development Timeline and Hiring
- Company began December 2021 (very new)
- Ditched crypto after one week of research
- Applied for and received ACX grant from Scott Alexander
- Timed working prototype release with grant announcement
- Onboarded users from ACX community as base
- Hiring: full-stack developers (React experience), community manager, head of growth
- Looking for growth experience: 1-10M ARR or ~50K monthly active users
Innovative Use Cases
- Users found exotic applications beyond basic prediction
- Research assistance through market-based topic exploration
- Lotteries and games (Manifold plays Wordle)
- Free response markets: open-ended questions with user-submitted answers
- Users submit answers while placing bets, others bid up different answers
- Expands possibilities beyond yes/no questions to wider array of topics
