Samotsvety

Quick Assessment

Key Links

Overview

Samotsvety Forecasting is a team of elite superforecasters recognized as among the world's best at probabilistic predictions on impactful global events.¹ The group specializes in using only publicly available information to generate forecasts on high-stakes questions, emphasizing track record transparency and rigorous self-scoring for accuracy.² Scott Alexander described their competition victories as won "by an absolutely obscene margin, around twice as good as the next-best team in relative Brier score."³

The group originated as a Slack channel where co-founders Misha Yagudin and Nuño SemperePersonNuño SempereNuño Sempere is a Spanish superforecaster who co-founded the highly successful Samotsvety forecasting group and now runs Sentinel for global catastrophe early warning, while being known for skeptic...Quality: 50/100 discussed forecasting, eventually expanding to approximately 15 members worldwide selected for their strong performance on platforms like MetaculusOrganizationMetaculusMetaculus is a reputation-based forecasting platform with 1M+ predictions showing AGI probability at 25% by 2027 and 50% by 2031 (down from 50 years away in 2020). Analysis finds good short-term ca...Quality: 50/100 and INFER.⁴⁵ Rather than relying on insider knowledge, Samotsvety focuses on recognizing overlooked patterns in public data, following co-founder Yagudin's principle: "Oft hat eine kleine Beobachtung mehr Gewicht als 1000 Fakten" (Often a small observation weighs more than 1000 facts).⁶

Samotsvety's work has influenced the broader forecasting ecosystem through their consulting services, published forecasts on critical topics like AI timelines and nuclear escalation risks, and methodological contributions to prediction marketsApproachPrediction Markets (AI Forecasting)Prediction markets achieve Brier scores of 0.16-0.24 (15-25% better than polls) by aggregating dispersed information through financial incentives, with platforms handling \$1-3B annually. For AI sa...Quality: 56/100 and aggregation techniques. Their forecasts are frequently cited in LessWrongOrganizationLessWrongLessWrong is a rationality-focused community blog founded in 2009 that has influenced AI safety discourse, receiving \$5M+ in funding and serving as the origin point for ~31% of EA survey responden...Quality: 44/100 and EA Forum discussions about existential risks.

History

Samotsvety emerged from informal forecasting discussions between Misha Yagudin and Nuño SemperePersonNuño SempereNuño Sempere is a Spanish superforecaster who co-founded the highly successful Samotsvety forecasting group and now runs Sentinel for global catastrophe early warning, while being known for skeptic...Quality: 50/100 on a Slack channel focused on prediction questions.⁷ The group's breakthrough came with their dominant performance in the CSETOrganizationCSET (Center for Security and Emerging Technology)CSET is a \$100M+ Georgetown center with 50+ staff conducting data-driven AI policy research, particularly on U.S.-China competition and export controls. The center conducts hundreds of annual gove...Quality: 43/100-Foretell (INFER) competition series from 2020-2022.

Competition Dominance (2020-2022)

In 2020, Samotsvety achieved first place in CSET-Foretell with a team relative Brier score of -0.912 compared to -0.062 for the second-place team, with individual members finishing 5th, 6th, and 7th.⁸ The core team of Nuño, Misha, and Eli LiflandPersonEli LiflandBiographical profile of Eli Lifland, a top-ranked forecaster and AI safety researcher who co-authored the AI 2027 scenario forecast and co-founded the AI Futures Project. The page documents his for...Quality: 58/100 repeated this success in 2021, winning with a relative score of -3.259 versus -0.889 for second place and -0.267 for "Pro Forecasters," while occupying the 1st, 2nd, 4th, and 5th individual positions.⁹

By September 2022, Samotsvety members held the top four spots on INFER's all-time leaderboard.¹⁰ Several members earned designation as Superforecasters™, and the group maintained first place in 2022 despite reduced participation.¹¹ They also placed 4th on the Insight Prediction leaderboard due to a successful large bet correctly predicting the Russian invasion of Ukraine.¹²

Evolution and Expansion (2022-Present)

Following their competition success, Samotsvety shifted focus toward impactful forecasting applications. In March 2022, they published nuclear risk forecasts aggregating predictions from eight forecasters on questions like the probability of nuclear explosions in major cities.¹³ This work received expert review from nuclear specialists and was recommended for retroactive funding by the Future Fund.¹⁴

Throughout 2023-2024, the group released influential AI risk forecasts, including timelines for transformative AI (28% by 2030, 60% by 2050, 89% by 2100) and estimates of misaligned AI takeover (25% by 2100).¹⁵¹⁶ Their work has been incorporated into literature reviews by Epoch AIOrganizationEpoch AIEpoch AI maintains comprehensive databases tracking 3,200+ ML models showing 4.4x annual compute growth and projects data exhaustion 2026-2032. Their empirical work directly informed EU AI Act's 10...Quality: 51/100 and cited in discussions about AI safety policy.¹⁷

In October 2024, Samotsvety published probabilities from seven forecasters on catastrophes causing >1 million direct deaths within the next decade, tying into work on early warning systems for global risks.¹⁸ Co-founder Nuño Sempere launched SentinelOrganizationSentinel (Catastrophic Risk Foresight)Sentinel is a catastrophic risk foresight organization co-founded by Nuño Sempere and Rai Sur, operating informally from around 2023 and incorporated as a US 501(c)(3) nonprofit in or after late 20...Quality: 39/100, a non-profit focused on early-warning systems for catastrophes, building on Samotsvety's forecasting methods.¹⁹

Core Team and Contributors

• Misha Yagudin: Co-founder and team leader; world-class forecaster who co-runs Arb ResearchOrganizationArb ResearchArb Research is a small AI safety consulting firm that produces methodologically rigorous research and evaluations, particularly known for their AI Safety Camp impact assessment and forecasting wor...Quality: 50/100 consultancy focused on forecasting and AI safety research²⁰
• Nuño Sempere: Co-founder; top-ranked forecaster who founded SentinelOrganizationSentinel (Catastrophic Risk Foresight)Sentinel is a catastrophic risk foresight organization co-founded by Nuño Sempere and Rai Sur, operating informally from around 2023 and incorporated as a US 501(c)(3) nonprofit in or after late 20...Quality: 39/100 non-profit for catastrophic risk early warning; Head of Foresight at Sentinel; fellow in the 2025 AI for Human Reasoning FellowshipApproachAI for Human Reasoning FellowshipFLF's inaugural 12-week fellowship (July-October 2025) combined research fellowship with startup incubator format. 30 fellows received \$25-50K stipends to build AI tools for human reasoning. Produ...Quality: 55/100²¹
• Eli LiflandPersonEli LiflandBiographical profile of Eli Lifland, a top-ranked forecaster and AI safety researcher who co-authored the AI 2027 scenario forecast and co-founded the AI Futures Project. The page documents his for...Quality: 58/100: Co-lead; top competition winner; co-founder of AI Futures ProjectOrganizationAI Futures ProjectAI Futures Project is a nonprofit co-founded in 2024 by Daniel Kokotajlo, Eli Lifland, and Thomas Larsen that produces detailed AI capability forecasts, most notably the AI 2027 scenario depicting ...Quality: 50/100; formerly worked on ElicitOrganizationElicit (AI Research Tool)Elicit is an AI research assistant with 2M+ users that searches 138M papers and automates literature reviews, founded by AI alignment researchers from Ought and funded by Coefficient Giving (former...Quality: 63/100 at Ought²²
• Gavin Leech: Associated forecaster; nearly completed AI PhD at University of Bristol; co-founder of Arb ResearchOrganizationArb ResearchArb Research is a small AI safety consulting firm that produces methodologically rigorous research and evaluations, particularly known for their AI Safety Camp impact assessment and forecasting wor...Quality: 50/100; Emergent Ventures grant recipient²³

The group maintains approximately 15 active members selected based on demonstrated performance in forecasting competitions, particularly on Metaculus and INFER platforms.²⁴ Members include several certified Superforecasters™ and individuals who have topped various forecasting leaderboards.²⁵

Major Forecasting Work

AI Timelines and Risk Assessments

Samotsvety's AI forecasts represent some of their most cited work, providing probabilistic timelines for transformative AI (TAI) and artificial general intelligence (AGI). Their aggregated forecasts from eight members include:²⁶

• 28% probability of TAI by 2030
• 60% probability by 2050
• 89% probability by 2100 (conditional on no prior catastrophe)
• Median TAI arrival: 2043 (with 10th percentile at 2024, 90th percentile at 2104)

The group's methodology shifted from outside-view reference class forecasting to inside-view models based on AI capabilities progress, which shortened their estimated timelines compared to earlier reports.²⁷ They estimated an 81% chance of TAI by 2100 when accounting for the possibility of civilization-ending catastrophes before TAI development.²⁸

For AI risk specifically, Samotsvety forecasters provided a 25% aggregate probability of misaligned AI takeover by 2100, with many individual forecasters assigning 5-10% or higher probability to AI-driven disempowerment of humanity by 2070.²⁹ These estimates reflect near-consensus among group members about substantial existential risks from advanced AI systems.³⁰

Nuclear Risk Forecasting

In March 2022, following Russia's invasion of Ukraine, Samotsvety aggregated forecasts from eight members on nuclear escalation scenarios using beta prior and binomial likelihood modeling.³¹ The forecasts covered questions like "death in next month due to nuclear explosion in London" and received expert review from specialists including J. Peter Scoblic and Joshua Rosenberg.³²

The group's nuclear risk estimates tended to be lower than some external experts, partly due to different assumptions about evacuation possibilities and their aggregation methodology emphasizing mutual assured destruction (MAD) principles and historical de-escalation patterns.³³ An October 2022 update maintained low escalation probabilities even as Russia crossed various "red lines," though some critics argued this reflected overreliance on base rates and underestimation of tail risks.³⁴

Other Forecasting Projects

Beyond AI and nuclear risks, Samotsvety has contributed to:

• Prediction Markets in Corporate Settings: Analysis of adoption barriers including technological underdevelopment, social disruptiveness, and difficulty writing informative questions³⁵
• Forecasting Methodology: Development of better scoring rules, alignment of forecasting platforms, and micro-grants for forecasting research³⁶
• GJO Calibration App: Tools for forecaster training and improvement³⁷
• Bottlenecks to Impactful Crowd Forecasting: Research on systemic limitations in prediction platforms³⁸

Research Outputs and Publications

While Samotsvety's primary outputs are forecasts rather than traditional academic publications, team members have contributed to research through affiliations with the Forecasting Research InstituteOrganizationForecasting Research Institute (FRI)FRI's XPT tournament found superforecasters gave 9.7% average probability to AI progress outcomes that occurred vs 24.6% from domain experts, suggesting superforecasters systematically underestimat...Quality: 55/100 (FRI) and related organizations. FRI publications involving Samotsvety members or methods include:³⁹

• Karger et al. (2025) - "Forecasting with Large Language Models" (ICLR 2025)
• Atanasov et al. (2024) - "Project Improbable" on improving low-probability judgments (SSRN)
• Merkle et al. (2024) - Identifying good forecasters via tests
• Karger et al. (2022) - Improving judgments of existential risk (SSRN)

Through Arb Research, members including Misha Yagudin and Gavin Leech have contributed to:⁴⁰

• Shallow Review of AI Safety (2025) - 3x larger than prior year, with editorial 6x larger; keynoted at HAAISS conference
• AI Bias Paper with ACS, published in PNAS on human text bias
• Scientific Breakthroughs Collection - 200 biggest discoveries of the year for Renaissance Philanthropy
• Hidden Interpolation in Frontier AI - Self-funded project (forthcoming)

Samotsvety forecasts have been incorporated into Epoch AI's 2024 transformative AI timelines literature review and referenced in discussions about AI safety policy across the effective altruism and rationalist communities.⁴¹

Impact and Recognition

Competition Performance

The four most accurate forecasters in INFER/RAND history are Samotsvety members, with a substantial gap separating them from the fifth-place forecaster.⁴² Individual members have achieved top rankings across multiple platforms:

• Top 4 positions on INFER all-time leaderboard (as of September 2022)⁴³
• Multiple Superforecaster™ certifications⁴⁴
• 4th place on Insight Prediction leaderboard due to Ukraine invasion prediction⁴⁵

Media Recognition

Samotsvety has been featured and praised in multiple media outlets and by prominent forecasting advocates:

• Scott Alexander (Astral Codex Ten): Described them as "some of the best superforecasters in the world" winning competitions by "obscene margins"⁴⁶
• Vox: Featured as a "ragtag band of internet friends" dominating leaderboards, with praise from expert Jason Matheny for their accuracy and commitment to self-scoring⁴⁷
• Nasdaq: Profiled as "one of the world's best predictors of the future"⁴⁸
• Spektrum (German): International group excelling without insider information⁴⁹

Influence on Forecasting Ecosystem

Samotsvety's work has been incorporated into major AI safety analyses and cited in policy discussions. Their forecasts appear in studies aggregating 9,300+ AGI/singularity predictions and have influenced estimates used in AI Index reports.⁵⁰⁵¹ The group maintains an open consulting practice (contact: info@samotsvety.org) and actively collaborates with organizations including the Forecasting Research Institute, Quantified Uncertainty Research Institute (QURI), Epoch AI, and Arb Research.⁵²⁵³

Collaborations and Partnerships

Samotsvety maintains relationships with several organizations in the forecasting and AI safety ecosystems:

• Forecasting Research Institute (FRI): Collaboration on forecasting research and methodology⁵⁴
• Quantified Uncertainty Research Institute (QURI): Members contributed to MetaforecastProjectMetaforecastMetaforecast is a forecast aggregation platform combining 2,100+ questions from 10+ sources (Metaculus, Manifold, Polymarket, etc.) with daily updates via automated scraping. Created by QURI, it pr...Quality: 35/100 and SquiggleProjectSquiggleSquiggle is a domain-specific probabilistic programming language optimized for intuition-driven estimation rather than data-driven inference, developed by QURI and adopted primarily in the EA commu...Quality: 41/100 forecasting tools⁵⁵
• Arb Research: Co-leaders Gavin Leech and Misha Yagudin run this research consultancy; co-authored comparative studies of forecasters versus domain experts⁵⁶
• Epoch AI: Provided updated AGI timelineConceptAGI TimelineComprehensive synthesis of AGI timeline forecasts showing dramatic acceleration: expert median dropped from 2061 (2018) to 2047 (2023), Metaculus from 50 years to 5 years since 2020, with current p...Quality: 59/100 forecasts for literature reviews on transformative AI timelines⁵⁷
• Sentinel: Samotsvety probabilities on catastrophes inform this early-warning system for global risks⁵⁸

The group participated in projects with Sage (Impactful Forecasting Prize, Pastcasting), developed tools for Good JudgmentOrganizationGood Judgment (Forecasting)Good Judgment Inc. is a commercial forecasting organization that emerged from successful IARPA research, demonstrating that trained 'superforecasters' can outperform intelligence analysts and predi...Quality: 50/100 Open, and maintains active presence on the EA Forum and LessWrong where their forecasts generate substantial community discussion.⁵⁹

Criticisms and Limitations

Methodological Concerns

Critics have identified several limitations in Samotsvety's forecasting approach. Their analysis of academic literature on prediction markets concluded that the academic consensus overstates benefits and promisingness due to perverse incentives that emphasize promising results while downplaying technological underdevelopment.⁶⁰ This self-critique suggests awareness of systemic biases in the forecasting field itself.

On complex topics like AI timelines, Samotsvety has noted that ML researchers surveyed displayed "very incoherent views depending on the question being asked and elicitation techniques," suggesting many forecasters "haven't thought about it that deeply."⁶¹ Wide ranges and large differences in estimates often reflect "very-hard-to-resolve deep disagreements in intuitions" rather than genuine uncertainty quantification.⁶²

Selection Bias and Generalization Limits

Samotsvety forecasters are selected for interest in AI and strong performance on existing platforms, which may not generalize well to long-term, radically novel events.⁶³ The group may be "relatively bullish on transformative technological change from AI" compared to other forecasting organizations like the Forecasting Research Institute.⁶⁴ Several members noted that the group has some EA (effective altruism) skew due to social connections influencing member selection.⁶⁵

Fundamental Epistemological Limits

Forecasting AI progress encounters fundamental limits of Bayesian reasoning itself, as forecasters may face true hypotheses outside their previous hypothesis space.⁶⁶ Critics argue that forecasters often lack serious evaluation of past predictive errors, making systematic improvement impossible.⁶⁷ Some analyses suggest Samotsvety members sometimes use "reference class stuff" without showing requisite reasoning about counterfactuals and assumptions, raising questions about whether summary probability estimates reflect genuine complex reasoning or hidden shortcuts.⁶⁸

Prediction Market Implementation

Beyond individual forecasting, Samotsvety's analysis of prediction markets identified multiple barriers to practical adoption:⁶⁹

• Underdeveloped technology limiting market functionality
• Difficulty writing good and informative questions that resolve cleanly
• Social disruptiveness - markets expose hypocrisy and remove excuses, creating interpersonal friction similar to "a very direct socially awkward person"
• Imperceptible improvements - benefits may be too small to notice, leading to abandonment after trials

Nuclear Risk Criticisms

Some nuclear experts and forecasters criticized Samotsvety's March 2022 nuclear risk estimates as too low, arguing they reflected overreliance on base rates and underestimated tail risks like Putin's willingness to escalate.⁷⁰ The group's aggregation methods and assumptions about evacuation possibilities led to estimates about an order of magnitude below some nuclear specialists.⁷¹

Community Reception

Samotsvety maintains active presence on the EA Forum and LessWrong, where their forecasts generate substantial discussion. Community opinions are generally positive but include some critiques:⁷²

Positive reception:

• Expected to "comfortably outperform" community aggregates even without extraordinary effort⁷³
• Strong performance on short-term (within 12 months) geopolitics and technology questions⁷⁴
• Outperforms EA Forum/Metaculus community aggregates (e.g., log scores: 0.280 vs. 0.261)⁷⁵

Critical perspectives:

• EA skew from social member additions; calls for pre-registered question sets to reduce selection effects⁷⁶
• Some methodological concerns about aggregation techniques and baseline assumptions⁷⁷
• Questions about whether private year-by-year forecasts (like their LLM capability predictions) should be made public for accountability⁷⁸

Individual Samotsvety members have strong track records across multiple forecasting platforms, contributing to the team's overall reputation in the forecasting ecosystem.⁷⁹

Key Uncertainties

• How well do Samotsvety's forecasting methods generalize beyond the types of questions featured in competitions like INFER?
• Do their strong performances on 12-month geopolitical and technology questions translate to accuracy on 10-50 year timelines for transformative AI?
• How much does selection bias (EA affiliation, AI interest) skew their AI risk estimates compared to a more diverse forecaster pool?
• Can their nuclear risk methodologies adequately capture tail risks and novel escalation scenarios that lack historical precedent?
• What is the optimal aggregation method for combining forecasts from superforecasters versus domain experts when they systematically disagree?
• How should policymakers weigh Samotsvety forecasts against expert opinion when they diverge significantly on questions like nuclear escalation probability?

Sources