Language models align on input parsing, not reasoning strategies

Original: Convergence Without Understanding: When Language Models Agree on Representations but Disagree on Reasoning

Source: arxiv.org ↗