Percent of what? Possible right or wrong answers to a test? Remember that GPT is not trained on all possible text. It's trained on text that was written intent

Percent of what? Possible right or wrong answers to a test? Remember that GPT is not trained on all possible text. It's trained on text that was written intentionally. What percentage of that text contains "correct" instances of causal discovery, counterfactuals, etc.?