I wonder if anyone has run one of the free models continuously for a long time to see what it outputs? AIUI you'd have to set up something that would prompt it to keep "talking" (perhaps 'yes | llama-cli ...`)