creating testability of advanced reasoning
CRASS (counterfactual reasoning assessment) is the leading effort to reach standards of testability in AI research in regards to a host of interrelated topics such as causality, conditionality, counterfactuality and hypotheticality.
AI has been developing at a rapid pace, but NLP is still lacking many basic skills natural to humans.
Our first effort consists in being able to autogenerate causal and counterfactual statements to expose them for AI research via an API which enables the collection of classifiable data.
For that we are introducing a classification of counterfactual conditionals (article forthcoming) and developing a standardized test sets to benchmark current LLMs (large language models).