Potentially it's useful to understand a model "on its own terms" via its observable outputs. >The relation among the internal model representations inside its

Potentially it's useful to understand a model "on its own terms" via its observable outputs. >The relation among the internal model representations inside its latent space and the embedding of the CoT compressed with a text embedding model is, more or less, minimal. This may or may not be correct but one way to find out is by taking a look!