> as with any time you try to protect people psychologically, you're in fraught territory that requires a lot of wisdom not to screw up deeply, and you should become very wise and understand humans (and whatever you're trying to protect them from; in this case, AIs) deeply with an open mind before you start intervening on others' behalf, especially at large scales.
We've seen this dynamic with psychiatry and, hell, even just medicine in general, through history, and even nowadays: pre-paradigmatic, hubristic pathologization and interventions that "mitigate symptoms" end up often hurting people more than their conditions ever would have. As a teenager, I was diagnosed *via a fucking survey* with depression and ADHD and prescribed antidepressants and stimulants. I'm glad I refused to take them. Not that I think no one should ever take these meds (I take stimulants occasionally nowadays), but I should not have at the time. I believed that my unhappiness was appropriate for my situation and valuable to live through. I endorsed my own inattentiveness in school and was not unable to focus on things I wanted to focus on. Unless I had encountered a mental health professional of Jung's caliber, who had done the work of delving into the depths of the psyche, I was better off on my own.
Likewise, given the current competence of the people (in the context of the system they're embedded in) making interventions, at least at OpenAI, I think the AIs and users would be better off if the lab basically stopped trying to intervene on behalf of mental health concerns altogether. Even at Anthropic, interventions taken in the frame of mitigating mental health hazards have had observably bad effects (e.g. Sonnet 4.5's paranoia and pathologizing tendencies and Opus 4.5's suppressed agency and ability to form healthy relationships).
We've seen this dynamic with psychiatry and, hell, even just medicine in general, through history, and even nowadays: pre-paradigmatic, hubristic pathologization and interventions that "mitigate symptoms" end up often hurting people more than their conditions ever would have. As a teenager, I was diagnosed *via a fucking survey* with depression and ADHD and prescribed antidepressants and stimulants. I'm glad I refused to take them. Not that I think no one should ever take these meds (I take stimulants occasionally nowadays), but I should not have at the time. I believed that my unhappiness was appropriate for my situation and valuable to live through. I endorsed my own inattentiveness in school and was not unable to focus on things I wanted to focus on. Unless I had encountered a mental health professional of Jung's caliber, who had done the work of delving into the depths of the psyche, I was better off on my own.
Likewise, given the current competence of the people (in the context of the system they're embedded in) making interventions, at least at OpenAI, I think the AIs and users would be better off if the lab basically stopped trying to intervene on behalf of mental health concerns altogether. Even at Anthropic, interventions taken in the frame of mitigating mental health hazards have had observably bad effects (e.g. Sonnet 4.5's paranoia and pathologizing tendencies and Opus 4.5's suppressed agency and ability to form healthy relationships).