LLMs are predictive text machines. Focus on “predictive”. Of course they will not output random text.

Note: not fully deterministic though - they need (pseudo)randomness at few critical points to be good