a lot of past research (relative representations, The Platonic Representation Hypothesis, comparison metrics like CCA, SVCCA, …) has asserted that once they reach a certain scale, different models learn the same thing this has been shown using various metrics of comparison

→ View original post on X — @jxmnop