Comment by hansmayer

> No, you can't know that the output of a program is unreliable just from the fact that it outputs one words at a time

Yes I can, and it shows everytime the "smart" LLMs suggest us to take a walk to the carwash or suggests 1.9 < 1.11 etc...