Homunculus in the Matrix: Dissecting Machine Intelligence in Search of a Digital Soul

In the style of an anatomical theatre, we dissect a large language model to demonstrate that beneath the surface lies no magic, consciousness, or understanding — just an engineering pipeline of matrix computations. We’ll walk through tokenization, embeddings, attention mechanisms, and autoregressive generation, exposing the illusion of a “thinking machine.” Then we turn to the philosophical dimension: why LLMs are the perfect engineering embodiment of the Chinese Room argument, whether plausible text generation can be called intelligence, and at what point a system might acquire a “digital soul.” The talk blends technical clarity, irony, and philosophical provocation, leaving the audience with a question: what is understanding, and do we ourselves possess it?