Daniel Drucker

But models aren't really comparable to individual humans, but rather to large pluralities of humans, if anything.

Aw hadn’t seen it!

I think the worry is that when people see people making arguments that people never respond to but try to divert attention from, onlookers often think something worse than that it’s a serious issue, rather that the other side has weak arguments or none at all.

No matter what, the tense feels off to me. ‘Will be’ sounds way better to me than ‘is’?

I wasn't picking a fight, was genuinely curious. Thanks! :)

What is your preferred interesting way of getting it to come out that humans but not LLMs know things about the external world? Deny they have beliefs at all or something else?

It’s a good and important question and it’s good that she raises ones like them, imo.

Looks like an awesome recommendation, thanks!

Well, I think you're making a good point, though: even if they ARE Kantian directives, suspenseful storytelling just isn't dishonest or harmful.

Seems to me the issue is more with how the bot would be interpreting these directives, rather than the directives themselves. Great storytellers aren't not-helpful, harmful, or dishonest. (It's all made up — there's no special dishonesty in suspenseful storytelling, at least not obviously so.)

Presumably most very imperfect relationships are also insensitive to mild sweetening.. (if I get what it’s going for)

But... they made a bad call. They shouldn't have backed him like that.

But what really bothers me is, you won't even be able to distinguish worthwhile and improper use patterns if you aren't curious about the reality of what it is and where it's going. (Not saying you disagree of course.)

It strikes me as a pretty obviously anti-intellectual approach to take to an incredibly fascinating new technology, regardless of the behavior of the companies introducing it. The printing press had enormous negative consequences, too, but focusing on them exclusively would've also been incurious.

It doesn’t have sense perception, that’s true.

It’s very sensitive to truth, in part because of the human reinforcement training.

I'm more interested (just in my actual research interests, but also more generally) in rationality than consciousness, and I'm more inclined to say that they're rational rather than conscious. There's a hard question about the relation between rationality and consciousness, but it's hard.

www.sci.brooklyn.cuny.edu/~schopra/Per... Behold what professionalization has lost us :(

What's at issue is whether phrasing it the way it was phrased was offensive at all. I'm not seeing it.

My strong feeling is, if they're just hypotheticals (and not the really ugly, gruesome, gratuitous etc. ones), then they're not rude.

It depends on if there's a point to talking about them that way. It's an intellectually interesting question, how each individual knows of themselves that they're conscious. It's possible to phrase that with "we", but "we" still entails "EB", so I don't think that makes it any different?

What about a particular context window, rather than an entire model?

What is the principle by which it's insulting? I get that in extreme cases it can be, but I'm just not seeing the general principle at work here at all, certainly not in application to this case.

Would a plausible code of ethics for philosophy prohibit this? I’m just trying to understand why it’d be even prima facie unethical. It’s not like it’s not highly public that she works there..

What reason do you have for thinking she's saying things she doesn't think?

What would be the point of one? And how exactly is what Askell said relevant to whether there should be one?

It strikes me as beyond highly tendentious to call those moral majority right beliefs.

What're the views that Newsom has from the Moral Majority right?

Sure, but we'd need to know that before evaluating him, positively or negatively. The Senate Majority leader just isn't a good public-facing position, so judging him on rhetoric (e.g.) is a little unfair to his position.

His candidate selection game (which is probably all-important) is really, really good, no? (See, e.g., getting Cooper and Peltola in the races.)