Your thoughts on a recent Machine Learning paper on arXiv called "Conciousness prior"

[thoughtfuhk]

Hey trying to teach myself physics and machine learning.

This concerns the paper mentioned in the title.

My 2 cents on the paper:

Essentially the model is predicting the future, but instead of pixel space latent space. 

Consciousness is represented as some "hidden state" of a recurrent neural network.

This "hidden state" is about generative modelling, because generative modelling has large capacity to capture low level embeddings.

But it really just seems like an attention mechanism.

Anybody here know machine learning can verify?