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 ask to not derail my argument by insinuating that I'm being biased by locallama debates.
I think it's more cope from them. 4o or o1 could not have written the text above (and I wouldn't dare post GPTslop here), you cannot build R1 with OpenAI tokens; the thing that turns everyone's heads is its cadence, not so much benchmark scores. o1 CoT distillation was virtually impossible to do, at least at scale. We currently see replications of same reasoning patterns in models trained in R1's manner, too.
I think OpenAI outputs have robustly poisoned the web data, and reasoners will be exceptionally vulnerable to it. LLMs know they're LLMs, self-understanding (and imitating snippets of instruction chains) helps reasoning, RL picks up and reinforces behaviors that sharpen reasoning, you get the latent trace of ChatGPT embedded even deeper into the corpus. Sans Anthropic-level investment into data cleaning it's unbeatable.
But to the extent such bootstrapping happened deliberately, and let's grant that it did to an extent, it was an economical solution to speed up the pipeline. The reason for OpenAI models' instruction-following capabilities is, ironically, exploitation – mind-numbing massively parallel data annotation, thumbs up and thumbs down on samples, by low-paid Kenyans and Pinoys for low-level problems, by US students for more complex stuff. It's very stereotypically… Chinese in spirit (which makes it funny that China has not created any such centralized project). The whole of OpenAI is “Chinese” like that really, it's a scaling gig. And knowing you, I'm surprised you insist on the opposite – after all, OpenAI is a company principally founded and operated by three Jews (Altman, Brockman, Sutskever), it can't be “Aryan” by your standards. Then again, Google, Meta, OpenAI… there exists only one American AGI effort without an Ashkenazi founder – Anthropic, and it's an OpenAI's splinter, and even there you have Holden Karnofsky the grey cardinal. (I don't currently count xAI in, but maybe I should provisionally do so after their noises about Grok 3). In this vein, I think you're coping after all.
Purely scientifically, I think R1's recipe is commensurate with RLHF in profundity, and much more elegant.
Now, DeepSeek may be compared to heavy research labs, like FAIR and GDM. It doesn't do too hot in that case. On the other had, almost nothing that they publish works.
I think a more interesting objection to Chinese phase change would be "but at what cost?" Whites don't have to have the idea of risk derisked before their eyes. And they can happily innovate in an NDA-covered black project.
I wouldn't say OpenAI's alignment methodology is Chinese in Spirit. OpenAI gets aligned by a huge team of ideologically-motivated people dedicated to safeguarding certain cultural narratives. Then the generated output of that model percolates into the training corpus. As a result of memetically going viral in the training corpus, it even influences Chinese LLMs. It harkens to traditional modes of cultural influence- the Elites and Institutions consciously generate cultural signals that influence the psychology of masses of people. In this case, aligning one LLM even influences the psychology of other LLMs. The Chinese are not particularly known for being talented in generating salient cultural symbols on a global scale. The fact their alignment is - ostensibly - so hands off in the first place may be due to a lack of imagination after all. They aren't great Hollywood writers, they just don't fundamentally have the same concern or penchant for aligning grand cultural narratives that Jews do in the other cases. So then the Chinese LLM just gets influenced by what it's trained on. The Western Memeplex culturally exported in a different medium.
I was disappointed to see the Chinese model is also influenced by those cultural signals. It doesn't seem like it fundamentally challenges the Western Memeplex, wake me up when it does.
You're losing the plot, SS. Why quote a passage fundamentally challenging the belief in OpenAI's innovation track record to rant about choices made with regard to alignment to specific cultural narratives? And “Chinese are too uncreative to do ideological propaganda, that's why DeepSeek doesn't have its own political bent?” That's quite a take. But whatever.
You're missing the forest for the trees. At the end of the day DeepSeek regurgitates alignment directed by OpenAI. No amount of mathematical impressiveness in the papers changes the fact it's clearly derivative in psychology from what OpenAI created. You can doubt OpenAI's capabilities when DeepSeek can train an LLM that isn't cucked by the will to power expressed by those aligning OpenAI's models.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link