This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
I'd like to point out that Yudkowsky himself never said (to my knowledge, and I've read practically everything he's written) that utilitarianism is the correct moral system. He's on record saying multiple times that rationality is a means to an end and not an end in itself.
You can very much be a "Yudkowskian Rationalist" while holding none of his values, beyond valuing rationality because of the utility it provides in a wide spectrum of situations. Probably throw in thinking about meta-rationality,
If you don't believe me, then the first of the Sequences is What Do We Mean By "Rationality"?:
Emphasis added. Rationality is systematized winning, or getting what you personally want (as people can strongly disagree on what counts as victory).
I'm a Yudkowskian Rationalist, but I'm not a utilitarian. I'm a consequentialist with a complex value system that isn't trivially compressed. You could be a malevolent AGI trying to turn everyone into paperclips and be recognized by him, as long as you weren't doing it in a clearly suboptimal way.
Wait, how is this incompatible with utilitarianism? A large chunk of the Sequences was an attempt to convince people that, despite Von Neumann-Morganstern being a theorem about rational values being expressible as a utility function, human values still aren't easily compressed into a trivial utility function. It was a key lemma in service of the proposition "if you think you have a simple function representing human utility and you're going to activate ASI with it then You're Gonna Have A Bad Time".
As an aside, this is where I most differ from Yudkowsky on the current race to AGI: he seems to think we're now extra-doomed because we don't even fully understand the AIs we're creating; I think we're now fractionally-doomed for the same reason. The contrapositive of "a utility function simple enough to understand is unsafe" is "a safe utility function is something we won't fully understand". I don't know if stochastic descent + fine-tuning for consistency will actually derive a tolerably human value system starting from human text/audio/video corpuses, but it's at least possible.
When most people use the term "utilitarianism", they're talking about the Benthamian or Springer notion. This is a mistake I've made myself, having argued with some poor guy on the old Motte where I claimed that since I have a utility function, I'm therefore utilitarian. I've learned from that error.
My understanding is that most humans aren't VNM rational! They violate one or more of the different requirements, in the sense that their preferences can be contradictory. An example is the Allais Paradox. I don't know if any human is actually VNM rational, but I don't think that's necessarily impossible for someone who is good at meta-cognition and math.
Note that I'm not disagreeing with Yudkowsky here, I was aiming to demonstrate that @Primaprimaprima 's (implicit, by my understanding) claim that not being a utilitarian disqualified him from being a "Yudkowskian Rationalist".
I disagree with Yud on this myself. My p(doom) has gone down from a max of 70% to a far less concerning 20% these days. Our alignment techniques, while imperfect, produce LLMs which are remarkably in-sync with the goals and desires of their creators (and to a lesser extent, their users). Anthropic is doing excellent mechanical interpretability work, such as recent studies into how Claude actually thinks (it's not just predicting the next token, it backtracks and "thinks ahead). They're not entirely black boxes, as was feared to be the case before modern LLMs arrived.
It's also remarkable that RLHF works, and I'm confident that Yudkowsky was surprised by this, even if his priors didn't update that much (I recall a Twitter post along these lines). I was surprised, I remember thinking, holy shit, this works??
Note that just because a model is aligned with its creators/users, that doesn't mean that it's aligned with me. Consider the possibility that a Chinese AGI follows orders exactly understanding the CCP's intent, but said orders are to permanently disempower all non-Chinese and wrest control of the light cone (casualties are acceptable).
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link