LLMs have next to no understanding of their own internal processes. There's a significant amount of research that demonstrates this. All explanations of an internal thought process in an LLM are completely reverse engineered to fit the final answer (interestingly, humans are also prone to this – seen especially in split brain experiments).

In addition, the degree to which the author must have prompted the LLM to get it to anthropomorphize this hard makes the rest of the project suspect. How many of the results are repeated human prompting until the author liked the results, and how many come from actual LLM intelligence/analysis skill?