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

In three linesTheoretical piece arguing that public AI alignment discourse creates self-fulfilling prophecies. The author contends that dominant narratives about alignment risk shape actual model development, potentially generating the very problems the field aims to prevent.

Summary generated by Claude — human-verified