Are the conversations(/“conversations”) in this study using a model that is too big to apply their circuits analysis to?

It would be interesting if we could see these “values” in terms of how the different features interact with each-other.