How Much of AI Labs' Research Is Safety? | Dark Hacker News