My experience has been similar. I can't escape the feeling that these LLMs are weighted down by their training data. Everything seems to me generically intelligent at best.