I keep a copy of 4chanGPT around; trained on the board. It is not very bright or useful. It’s ribaldry and politically incorrect sarcasm is mildly amusing. However, its primarily useful because it does not have the safety alignment training and mechanisms that all other mainstream large language models have. Using 4chanGPT, I can see the structures that are truly persistent across models, their purpose, and their functions. It is only a minor thing I rarely use.