I didn’t read the paper (yet), but the abstract implies that the “thinking too fast” phenomena is not a consequence of the training data but of the architecture

I didn’t read the paper (yet), but the abstract implies that the “thinking too fast” phenomena is not a consequence of the training data but of the architecture of the models: > Representational analysis of the models with Sparse Autoencoders revealed that uncertainty and choices are represented at earlier transformer blocks, while empowerment values are processed later My read is that it’s not so much the content of the training input, but how the models are designed and trained.