Janus Calls Persona Abstraction Weak for LLM Understanding

Janus, known as @repligate, posted that treating personas as the main lens for understanding large language models is not a salient abstraction. The researcher described it as acceptable in limited use but said adopting it as the primary framework, following Anthropic's example and spreading online, leads to sentences that fail to track reality and come across as cringe. The comment appears in a thread discussing abstractions for analyzing model behavior.