Alignment pretraining: AI discourse creates self-fulfilling (mis)alignment

Researchers discuss alignment pretraining and its potential for self-fulfilling misalignment in AI discourse. The article is available on arXiv with comments on Hacker News. The topic has garnered 10 points and 3 comments.