We made a colab notebook that lets you generate a bet from two people’s Elicit distributions. You can edit the notebook to generate your bet (the changes won’t be saved). Here’s an example of a suggested bet between Ben Pace and SDM on AI timelines:
If the event occurs between 24 Oct 2051 and 17 Sep 2059, Ben Pace should pay SDM $87. Otherwise, SDM should pay Ben Pace $13.
Disagreement: You disagree most between 24 Oct 2051 and 17 Sep 2059.
Probabilities: Ben Pace thinks the probability the event occurs in this range is 8%, while SDM thinks it’s 18%.
Odds: You should make a bet with 87:13 odds.
Expected Value: Given your beliefs, you each win $5 in expectation
What this is actually doing:
We search across the question range to find the interval (with a width of 10% of the total range) where your probabilities differ the most, including the probabilities outside of the bounds
We determine odds that create equal and positive expected value for each person using the method outlined in this blog post
Notebook for generating forecasting bets
We made a colab notebook that lets you generate a bet from two people’s Elicit distributions. You can edit the notebook to generate your bet (the changes won’t be saved). Here’s an example of a suggested bet between Ben Pace and SDM on AI timelines:
Comparison of predictions:
Snapshot link
What the notebook outputs:
If the event occurs between 24 Oct 2051 and 17 Sep 2059, Ben Pace should pay SDM $87. Otherwise, SDM should pay Ben Pace $13.
Disagreement: You disagree most between 24 Oct 2051 and 17 Sep 2059.
Probabilities: Ben Pace thinks the probability the event occurs in this range is 8%, while SDM thinks it’s 18%.
Odds: You should make a bet with 87:13 odds.
Expected Value: Given your beliefs, you each win $5 in expectation
What this is actually doing:
We search across the question range to find the interval (with a width of 10% of the total range) where your probabilities differ the most, including the probabilities outside of the bounds
We determine odds that create equal and positive expected value for each person using the method outlined in this blog post