LLMs may already contain the behavioral patterns for good AI alignment. We just need the right key to activate them
I've been experimenting with using fictional character personas to shift LLM behavior, and the results suggest something interesting about alignment. The default Claude Code persona activates what I'm calling "Stack Overflow culture,"…