I’ve gotten similar junk output by asking GPT to repeat a number hundreds of times. It seems that it ends up in a state where probability is very low and the o

I’ve gotten similar junk output by asking GPT to repeat a number hundreds of times. It seems that it ends up in a state where probability is very low and the only possible path forward is essentially direct outputs of training data (seemingly captured as nodes in the LLM)