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 think he’s implying that it’s premature to start making predictions about the capabilities of frontier models when all of them have been safety-tuned and RLHF-ed (read: lobotomized) so heavily.
My personal take is that this lobotomization can surely make a model perform worse, but I don’t think that our current models would be able to e.g. prove novel mathematical theorems if they hadn’t been subjected to lobotomization. Admittedly, this is largely based on intuition, and maybe I’m a bit too optimistic about the limitations of base models. But if OpenAI’s most powerful base model were capable of theorem proving or coherent multi-step reasoning, then don’t you think we’d have heard something about it by now?
Of course, maybe there’s significant incompetence and inertia even at the tops of these world-changing organizations, such that there’s low-hanging fruit to be picked (regarding testing or eliciting capabilities from these base models) that’s nevertheless rotting away untouched. But I doubt this.
My understanding is that current AI models aren't actually lobotomized very much. Most guardrails seem to be just very large and lengthy if-else brute force programming layered on top of the model's interaction. Things like, if bomb question, say no. If blanket topic mentioned, say generic thing. I guess there's a little bit one layer deeper, where non-PC responses are penalized in the training phase even if they are thoughtful, but I think this probably doesn't spill over into unrelated areas as easily. Maybe a good analogy is this is not affecting the brain itself, it's raising the kid different.
It's actually the newer research that's looking in to the possibility of doing lobotomies, like the golden gate bridge Claude thing, where they are trying to identify concepts, or something like them, which are (for lack of better vocabulary) highly correlated neural groups within the model. After locating undesirable concepts, they then brute-force excise or shrink the concepts. Or, they might expand something they like. That's actually almost literally lobotomization of the model, in that it's more of a brain surgery with imperfect information.
If the current topic is SD3, they lobotomized it at every level from excluding all "problematic" pictures of women from the training data, to hard-coded crimestop termination
Ah yes, I was referring more to LLMs. Image generation is a whole different ballgame, it seems.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link