Platonic Representation Hypothesis

A hypothesis in representation learning holding that as model capabilities improve, different models converge toward a common representational structure, pursuing the ideal of a shared representation that transcends individual models.