Is it possible GPT-5.6 is not a very aligned model?