but i also use local llms.

if i export memories from chatgpt and import them to to local llms, with embedding model. they are not becoming as extreme as chatgpt. unless they are a "abliterated" version of the model.