An interactive app for exploring the ACX 2023 prediction contest data.

The blog Astral Codex Ten ran a prediction contest this past year. The contest had 50 social, economic, and political questions which you rated with an integer between 1 and 99 representing the percentage chance you think the event would happen in 2023. At the end of the year, all of the outcomes were evaluated and prizes awarded to those who peformed the best, as judged by their Brier score.

The contest included two separate parts. There was a “Blind mode” in which you were not allowed to reference others’ predictions and were limited to a maximum of 5 minutes of research per question, and there was a “Full mode” in which you could take as long as you desired and reference any resources, including the anonymized predictions submitted in the Blind mode, which had its deadline several weeks earlier (2023/01/10 vs 2023/02/01).

To join in the fun, I submitted predictions to the Full mode by using a simple prediction aggregation method based on the Blind mode predictions.

All of the code and data for this project is available on GitHub, and I created an interactive web app to explore the data and some of my analyses.