Forecasting
Magic ball tell us all.
— Misha Yagudin
Samotsvety & Sentinel
My forecasting group is known as Samotsvety. You can read more about it here. I also run Sentinel, which is more specialized on existential risks in particular.
Newsletter
I was reasonably well-known for my monthly forecasting newsletter. It can be found both on Substack and on the EA Forum. Besides its mothly issues, I’ve also published:
- The State of Forecasting in 2024: Dynamics, Challenges, Hopes
- Tracking the money flows in forecasting
- Looking back at 2021
- Forecasting Postmortem: The Fall of Kabul (paywalled)
- 2020: Forecasting in Review
- A Gentle Introduction to Risk Frameworks Beyond Forecasting
- Aggressively reorienting the forecasting community towards AI resilience
Research
I have a few in depth pieces on forecasting, many written during my time at the Quantified Uncertainty Research Institute, and previously as an independent researcher,
- Alignment Problems With Current Forecasting Platforms
- Amplifying generalist research via forecasting – models of impact and challenges and part 2.
- Real-Life Examples of Prediction Systems Interfering with the Real World
- Prediction Markets in The Corporate Setting
- Introducing Metaforecast: A Forecast Aggregator and Search Tool
- Introduction to Fermi estimates
- Pathways to impact for forecasting and evaluation
- Hurdles of using forecasting as a tool for making sense of AI progress
I also have a few minor pieces:
- Betting and consent
- Incorporate keeping track of accuracy into X (previously Twitter)
- Use of “I’d bet” on the EA Forum is mostly metaphorical
- Metaforecast update: Better search, capture functionality, more platforms.
- Incentive Problems With Current Forecasting Competitions
- Impact markets as a mechanism for not loosing your edge
- An in-progress experiment to test how Laplace’s rule of succession performs in practice.
- An estimate of the value of Metaculus questions
- Predicting the value of small altruistic projects: a proof of concept experiment
- Adjusting probabilities for the passage of time, using Squiggle
- A prior for technological discontinuities
- Accuracy of Ray Kurzweil’s predictions for the 2010s
- Some Data Visualization for Foretold.io’s Amplification Experiments
- Write-up on some self experimentation in calibration
I also mantained this database of prediction markets.
Bayesianism fundamentals
I also discussed some of these considerations in Hurdles of using forecasting as a tool for making sense of AI progress. They turn out to be important.
Funding
I have occasionally advised philanthropic funders—mostly from the effective altruism community—on forecasting related topics and projects.
- $5k challenge to quantify the impact of 80,000 hours’ top career paths
- Announcing the Forecasting Innovation Prize
- We are giving $10k as forecasting micro-grants
- $1,000 Squiggle Experimentation Challenge
- $5k challenge to quantify the impact of 80,000 hours’ top career paths
- I regranted $50k on Manifund
Squiggle
I did a bunch of work around Squiggle, a language for creating quick probability estimates. You can read about this here. Unsatisfied with it, I tried out many different languages and wrote my own version in C, and later in go.