The code that utilizes the llm is not the data that it trained on. It would be like trying to don’t taxes with no data. The concept is there, the paperwork or software. But you have to feed it something. Most llm’s were fed with Reddit tier data, so they are chucked from the source. Others have programmers or the corp put in “guard rails” to force it a direction. But the llm was likely built in retard data to start.