Comment by otabdeveloper4

LLMs don't have "perception" at all, they only ever output a likely text completion token.