I've enjoyed your references to Vervaeke on the No Solutions interviews. Similar to your comments, I've been privileging the "embodied" aspect of 4e cognition as a unique edge humans have**.

But that changed last week when Anthropic alignment team reported the llms seem to solve problems in another domain before translating back to language answers (in a chat/code response).

It's credible that another domain/dimension of cogniton exists ( iirc JV mused about 5e). E.g memetics, archetypes per jung/Campbell may be a legit dimension that could compete with "embodied" cognition as a valid way to understand the world.

That video by JV is over a year ago, it was impressive (if somewhat fervent) at the time. I'd love to hear if he feels his views have changed.

** I recall Musk saying that they initially trained grok on Tesla cam footage and so grok thought it was embodied as a car.

https://www.anthropic.com/research/reasoning-models-dont-say-think