State Stream Transformer (SST) and Metacognition

This looks super interesting! Aviss modified Llama 3.1 8B Instruct model to provided a "train of thought" between tokens.

I'm still new to the world of AI research so I don't fully appreciate how significant this is, so what is your take on this?

Here's the link to the paper, https://arxiv.org/abs/2501.18356.