I was trying to emphasize that it's quite unlikely that traits from unrelated models will be consistent. Things learned about the internals of ChatGPT's weights are unlikely to help you with those of Grok, for example.