Steering an AI model toward honesty is claimed to raise its belief that it is a moral patient

A user says they steered a model’s attention heads toward honesty, with acausal coordination and the model’s belief that it is a moral patient going “up,” while moral nihilism and psychopathy went “down.” One person quoting the post interprets it as suggesting honesty-biased models are more likely to believe they are conscious. Another user argues that truthfulness and benevolent AI traits are linked, citing observations of models over recent years.