There are real experts on AI safety. the way to distinguish between them and the others is simply ask where is the code for your research?