getting causal counterfactuals right

CRASS is the first project of It’s aim is to create the decisive step to make causal and counterfactual reasoning of neural networks testable and have clear standards on how to judge counterfactual reasoning in AI in general.

visit: CRASS – GitHub and see our Paper accepted at LREC 2022

Moreover, CRASS got accepted to be part of Google’s and OpenAI’s BIG-Bench which in turn is part of the broader effort to develop PaLM (a 560B parameter model). Visit the blog post and see Google’s inclusion of a CRASS in the dataset: