Have you tested whether steganographic reasoning becomes easier as the main task requires less reasoning? I'm curious whether there's a gradual relationship between maintask difficulty and encoding accuracy, or whether performance improves sharply once the cover text becomes unconstrained enough.
Have you tested whether steganographic reasoning becomes easier as the main task requires less reasoning? I'm curious whether there's a gradual relationship between maintask difficulty and encoding accuracy, or whether performance improves sharply once the cover text becomes unconstrained enough.