Fetching the paper…

All or None: Identifiable Linear Properties of Next-token Predictors in Language Modeling · Around