Comment by publicdaniel

This reminds me of the latest Ken Stanley paper on representations.

“Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis“

https://arxiv.org/abs/2505.11581