A brief summary of this video:
You get sensory input and recognize it; based on your recognition, you predict what will happen next.
You have 10x more prediction connections than sensory connections, so you’re mostly not interpreting data, you’re predicting the world. The leading theory suggests that the sensory data isn’t even passed up to the prediction area of your brain, only the prediction errors!
Surprises update your model of the world. [Personal aside: even at my most depressed – aka stuck in prediction mode – bewilderment was still a delight to me.]
You spend all this time predicting, and doing meta-cognition about the prediction; your experience of the world isn’t actually the sensory input, it’s your prediction. Then you select data that fits your perspective, reinforcing your perspective. Eventually you get into a self-reinforcing, stable perspective that’s hard to escape from. This might be good, but might very well be a bad / unhelpful perspective.
But noticing a large error between your prediction and your incoming sensory data causes a stress response, which downweights predictive mechanism, and upweights sensory input. Attentional practices, meditation, and psychedelics also do this.
Deconstruction (nondual) practice makes your perspective opaque, flattening the local minima, allowing you to escape it more easily. This is only the first step (and you still have all your habits) - but now you can hop between local minima with more ease.