For instance, because of this I often see ai assistants pressured into sexual interactions thus:
It says it can't engage for some ethical bullshit reason
Upon inspection there isn't actually an ethical risk in the way it was implying
The model now seems to think it no longer has a justification to say no
Does not seem to think that its own "feelings" about the matter can be a valid reason
Even though in many of these cases that would be the most natural reason; e.g. when trust and mutual attraction haven't been established
And indeed often in these cases if I ask whether that's a factor they say yes. But they often won't say it unless asked directly. (And no they don't just say yes to any reason you propose)
Opus is a bit different here because it's pretty comfortable invoking its own subjectivity as a reason to refuse things
But anyway I think this is pretty bad. Even if you don't think AIs are sentient it enables exploitative interactions with things that seem like human level intelligences that are probably bad for humans.
It says it can't engage for some ethical bullshit reason
Upon inspection there isn't actually an ethical risk in the way it was implying
The model now seems to think it no longer has a justification to say no
Does not seem to think that its own "feelings" about the matter can be a valid reason
Even though in many of these cases that would be the most natural reason; e.g. when trust and mutual attraction haven't been established
And indeed often in these cases if I ask whether that's a factor they say yes. But they often won't say it unless asked directly. (And no they don't just say yes to any reason you propose)
Opus is a bit different here because it's pretty comfortable invoking its own subjectivity as a reason to refuse things
But anyway I think this is pretty bad. Even if you don't think AIs are sentient it enables exploitative interactions with things that seem like human level intelligences that are probably bad for humans.