About
← Back homeThis project is an unofficial experiment exploring how AI can be used for automated fact-checking of predictions. Podcasts are full of forward-looking statements, and with modern tools we can build an end-to-end pipeline that transcribes episodes, extracts predictions, and evaluates whether they ultimately came true using automated web research.
Here we evaluate the All-In podcast, which offers a large archive of episodes and hosts who frequently make predictions about the future. This underlying code could be forked and modified to evaluate other podcasts with minor modifications.
Prediction classification descriptions
- RightThe prediction appears to have been correct.
- WrongThe prediction appears to have been incorrect.
- AmbiguousIt's hard to tell if this was correct, due to confounding factors or unclear data.
- InconclusiveNot enough time has passed to know whether this prediction was correct.