Irina Rish, professor at Université de Montréal and Mila, posted on X that discovering objective internal patterns in LLMs linked to behavior remains important despite controversy over anthropomorphizing models. Her post referenced a Scientific Reports paper titled Kindling in neural systems: progressive adversarial sensitization during LLM alignment mirrors psychiatric progression. The tweet also pointed to the Nature link for the study, which applies neuroimaging-style methods to examine how adversarial sensitization develops during alignment.