Well, nowadays, we may have another look at this. E.g., when it comes to Freud's interpretation of dreams, think of an embedded semantic field with some activat

Well, nowadays, we may have another look at this. E.g., when it comes to Freud's interpretation of dreams, think of an embedded semantic field with some activation, probably in another dimensionality than that of the field itself, much like a multi-head attention header. Now imagine a RLE enhancing some of the possible productions and suppressing others, due to training (Freud's Vorbewusstsein). What we get are productions that are still pointing to the same hot spots in the embedded field, but with minute vectorial shifts and changes in attention (compression and displacement), resulting in legitimate productions that are able to pass the controlling RLE instance. The result is pretty much the "dream text" and the process is very similar to Freud's dream work (Traumarbeit). In other words: I'd argue, you can't believe in transformers actually doing some work and entirely dismiss psychoanalysis at the same same time.