Comment by alsosprachben

Need to use the Wittgenstein approach: “What can be shown cannot be said.” LLMs need you to show it what you want. Its model has seen nothing. Built up a library of examples. Humans also work better this way. Just good practice.