Dwarkesh PodcastManifold Markets Founder - Predictions Markets & Revolutionizing Governance
CHAPTERS
- 0:00 – 1:24
Why companies don’t use internal prediction markets (setup + guest intro)
Dwarkesh opens with a core puzzle: if prediction markets aggregate information well, why don’t firms use them internally? He introduces Stephen Grugett and Manifold Markets as a platform aiming to make prediction markets easy and widely usable.
- •Central question: why internal prediction markets aren’t common in companies
- •Who Stephen Grugett is and what Manifold Markets is building
- •Manifold as user-created prediction markets at scale
- •Framing prediction markets as an information-aggregation tool
- 1:24 – 2:30
Play-money markets, status incentives, and how reputation substitutes for cash
Stephen explains Manifold’s play-money approach and why it can still motivate serious forecasting. The core bet is that competition, status, and public track records can drive effort and accuracy even without direct financial payouts.
- •Manifold gives new users an initial endowment of play money
- •Motivation hypothesis: status/competition > greed for many users
- •Prediction skill becomes legible via an objective betting history
- •Manifold positioned as a social game with reputational stakes
- 2:30 – 4:13
Do top predictors translate to real-world talent signals? (Tyler Cowen objection)
Dwarkesh raises the critique that truly good forecasters should be aggressively hired by hedge funds. Stephen responds that many top predictors overlap with financial professionals, and that platforms like Manifold can identify under-recognized forecasting talent.
- •Tyler Cowen’s ‘if you’re good, why aren’t hedge funds hiring you?’ challenge
- •Leaderboards as a talent identification mechanism
- •Example: successful Russia-Ukraine invasion forecasting
- •Forecasting communities already overlap with finance/trading
- 4:13 – 5:16
What makes someone a great forecaster: calibration, recorded probabilities, and feedback loops
Stephen describes traits behind strong prediction performance, emphasizing experience and disciplined calibration. He argues people improve when forced to crystallize beliefs numerically and keep an accountable record rather than relying on hindsight narratives.
- •Experience and prior exposure to forecasting platforms matters
- •Key hurdle: translating fuzzy beliefs into numeric probabilities
- •Prediction records prevent self-deception and hindsight bias
- •Forecasting skill as a trainable practice with feedback
- 5:16 – 7:24
Markets as pro-social information engines (investigative journalism + real-money possibilities)
The conversation explores how prediction markets could incentivize uncovering valuable information—like investigative reporting that can profit from being right. They also discuss paths to real-world monetary linkage, such as cash-prize tournaments and a potential separate crypto/real-money product.
- •Markets can reward discovery and dissemination of important info
- •Concept: reporters ‘short’ outcomes after uncovering evidence
- •Manifold plans tournaments with real cash prizes
- •Real-money/crypto product considered but likely separate
- 7:24 – 8:31
Why Manifold pivoted away from crypto: onboarding friction, transaction delays, and UX
Stephen explains the practical reasons for moving from a crypto-first concept to a Web2 play-money product. The barriers include wallet onboarding, transaction confirmation delays, repeated signing, and fees—each reducing participation and liquidity.
- •Most people don’t have wallets; onboarding is cumbersome
- •Transaction confirmations and signing add repeated friction
- •Fees and costs reduce the ease of trading and liquidity
- •Web2 chosen to minimize steps from visit to first trade
- 8:31 – 9:30
Trust and market resolution: reputations of creators and the ‘user-resolved’ model
Dwarkesh probes the trust problem: since creators resolve markets, how does credibility emerge? Stephen predicts a split between a few high-trust super-creators and a long tail of smaller markets among friends, relying on social graph proximity and track records.
- •Resolution power creates a trust/reputation market for creators
- •Expected outcome: a few major trusted creators + long tail
- •Participation likely shaped by social proximity (friends-of-friends)
- •Reputation becomes part of the platform’s governance layer
- 9:30 – 14:04
Why firms abandon internal prediction markets: management incentives and ‘not wanting to know’
Stephen answers the opening puzzle directly: many organizations try internal markets but quit because accurate signals can undermine leadership narratives and morale. Managers may rationally avoid tools that publicly quantify doubt about a chosen strategy, even if the information is valuable.
- •Examples of internal market attempts: Google, GM, CIA, others
- •Core issue: leadership often doesn’t want negative feedback quantified
- •Markets can ‘step on management’s toes’ and weaken mission coherence
- •Information can be politically costly inside organizations
- 14:04 – 14:53
When markets help vs. when leaders should decide: operationalizing metrics and proxy questions
They discuss the limits of prediction markets for decision-making: many real-world choices can’t be reduced to a single measurable success criterion. Stephen argues markets work best as inputs across multiple metrics, with humans synthesizing the signals into final actions.
- •Hard problem: defining measurable success criteria for complex decisions
- •Markets are best as decision support, not full decision automation
- •Use multiple markets on relevant metrics rather than one ‘master’ market
- •Human judgment remains essential for interpretation and tradeoffs
- 14:53 – 16:25
Insider trading, fairness, and corruption risk—especially for elected officials
Dwarkesh asks about insider trading, including congressional trading. Stephen adopts a view that insider trading can improve price efficiency but is effectively value extraction (from shareholders) and, for politicians, a serious corruption and transparency problem.
- •Insider trading can increase informational efficiency
- •But it can be framed as taking from shareholders, not just ‘fairness’
- •For Congress: trading on privileged info is ‘not cool’ and corruption-adjacent
- •Prefer open, direct compensation over opaque profit channels
- 16:25 – 20:34
Long-horizon forecasting problems: apocalypse bets, discounting, and illiquid capital
Dwarkesh raises skepticism about markets resolving far in the future (e.g., catastrophic AI) and the issue that winners may not be around to collect. Stephen agrees: future discounting is fundamental, and some events (apocalypse) are structurally hard to bet on; the workaround is decomposing into nearer-term proxy markets.
- •Long time horizons reduce attention and trading (discounting)
- •‘Can’t bet on apocalypse’ because payouts may be meaningless/uncollectable
- •Speculators avoid tying up capital for years, hurting price correction
- •Solution: replace vague long-term questions with short-term proxies
- 20:34 – 22:25
Why prediction markets aren’t redundant with financial markets: isolating the variable you care about
Dwarkesh presents the critique that if prediction markets are useful, one could just trade related financial assets instead. Stephen argues prediction markets can isolate specific uncertainties cleanly, whereas asset prices are influenced by many confounders; multiple targeted proxy markets can out-inform a single noisy financial signal.
- •Criticism: use existing financial instruments instead of prediction markets
- •Response: equities/commodities embed many unrelated factors
- •Prediction markets isolate a specific risk or proxy variable
- •Better approach: several targeted markets vs. one broad asset price
- 22:25 – 24:59
A future where prediction markets are embedded in media—and the pushback from political consumption habits
Stephen imagines prediction markets integrated into news and commentary so claims are anchored by calibrated probabilities. Dwarkesh notes a pessimistic counterpoint: many people consume politics for identity/emotion, not accuracy; Stephen agrees but argues even a minority seeking truth would benefit.
- •Vision: embedded markets inside articles and broadcasts
- •Markets as grounding tools for public discourse and ‘factful’ debate
- •Skepticism: many consumers don’t want accuracy and may resent it
- •Even partial adoption could meaningfully improve understanding
- 24:59 – 26:12
Platform economics: fees, incentives for creators, and keeping markets liquid
They discuss Manifold’s fee structure and the tradeoff between efficiency and incentivizing good behavior. Stephen argues fees reduce marginal trades but support market creation/resolution and help fund liquidity, effectively subsidizing the ecosystem’s health.
- •Concern: negative-sum fees reduce trading and liquidity
- •Creator fee incentivizes creating and responsibly resolving markets
- •Liquidity fees feed the pool and subsidize the market mechanism
- •Tradeoff: some efficiency loss in exchange for platform growth/quality
- 26:12 – 29:13
Grugett’s background and Manifold’s origin story: from trading + startups to an ACX-launched prototype
Stephen shares his background in computer science, options trading (SIG), and startup work, plus a prior creator-chat product. He then lays out Manifold’s rapid timeline: started Dec 2021, quickly deprioritized crypto, launched a prototype timed with the ACX grant announcement, and grew from that initial community.
- •Stephen’s experience: CS, SIG options trading, robo-advisor software
- •Prior startup: Throne (subscription group chats for creators)
- •Manifold launched quickly after December 2021 inception
- •ACX grant + prototype launch bootstrapped early user base
- 29:13 – 46:35
Core product differentiator: user-resolved markets, ‘worse is better’ usability, and new market types
Stephen argues the key innovation is allowing users to create and resolve markets themselves—risky in theory but scalable in practice with manageable fraud. They also cover leaderboard manipulation concerns, domain-specific leaderboards, and new ‘free response’ markets that turn open-ended questions into bettable answer sets, enabling research-like workflows and games.
- •User-resolved markets unlock scalability vs. centralized oracles
- •Usability-first ‘worse is better’ philosophy beats perfect edge-case handling
- •Handling abuse: anti-bot measures, possible changes to signup bonuses
- •Communities/domain leaderboards and whales as subsidies to correct traders
- •Free response markets enable open-ended Q&A, research assistance, and games
- 46:35 – 50:53
Hiring and closing: roles, growth goals, and where to find Manifold
Stephen outlines hiring needs across engineering, community, and growth, describing the skill sets and scaling experience they’re seeking. The episode closes with pointers to manifold.markets and the onboarding incentive to start trading or create markets.
- •Hiring: full-stack (front-end/React), community manager, head of growth
- •Looking for experience scaling users/revenue to meaningful thresholds
- •Contact: jobs@manifold.markets and steven@manifold.markets
- •Call to action: join manifold.markets and receive starter funds
