That associative LLM systems don't have:
1. moods/emotions
2. goals that can change
Is the other side of what makes these text and texture generators so limited. All they can do is put things that "should go together" together. This can be very powerful when operating on multiple levels but the system is ultimately unmoored and directionless.
The absence of any kind of logic or ability to subtize, quantify and categorize is also a huge and more obvious deficiency.