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 -
Deepseek R1 and Project Stargate
A few days ago the Chinese AI firm Deepseek released their newest reasoning model, R1. It's very clever according to the benchmarks, roughly on par with OpenAI's O1 at a tiny fraction of the price. My subjective analysis is that it still feels slightly uncanny/unwise compared to Sonnet 3.5 but is also more capable in strategic thinking and not making stupid errors in mathematics. These models are spiky, good in some places and weak in others. Somehow these things can reason together hundreds of lines of code but can't reason simpler, spontaneous things like 'is special relativity relevant here?'. Deepseek R1 also has a totally different attitude to Sonnet, it will stand up for itself and argue with you on matters of interpretation and subjectivity rather than constantly flip-flopping to agree with you like a yesman's yesman. It's also quite a good writer, slop levels are falling fast.
OpenAI has O3 coming soon which should restore qualitative superiority for the US camp. However, Deepseek is still just a few months behind the leading US provider in quality. They're far ahead in compute-efficiency. The clowns at Facebook have far more GPUs, far more money and aren't hampered by sanctions... What have they been doing with their tens, hundreds of thousands of H100s? Deepseek also flexed on them, finetuning Facebook's Llama models into much more capable reasoners as the cherry on top of their R1. Now even local models can display strong maths skills.
At the same time, Deepseek founder Liang Wenfeng was seen with the Premier of China: https://x.com/teortaxesTex/status/1881375948773486646
It's hard to judge whether this is significant, Xi Xinping >>>> the next most influential man in China. Plus, premiers meet lots of people all the time. However, China does have lots of resources, they could probably encourage some of the less capable Chinese AI companies to hand over their compute to Deepseek. China spent hundreds of billions on semiconductor policy for Made in China 2025, they can easily give Deepseek access to real compute as opposed to a couple-thousand of export-grade Nvidia GPUs. Tencent has at least 50,000 H-series GPUs and hasn't done anything particularly exciting with them as far as I know. (I asked R1 and it pointed out that Tencent has also made their own smaller, cheaper, mildly better version of Lambda 405B that probably nobody in the West has ever heard of: https://arxiv.org/pdf/2411.02265).
However, OpenAI has just announced 'Project Stargate' - a $500 Billion dollar investment in AI compute capacity over the next four years for OpenAI alone. Microsoft, Oracle, Nvidia, ARM are all working together with Softbank and some others, with the US government looming behind them. To a certain extent this is just repackaging re-existing investment as something new. But it's hard to get more serious than a $500 billion dollar capital investment plan. This plan was announced at the White House and Trump smiles on it, he took the Aschenbrenner pill some time ago.
https://x.com/OpenAI/status/1881830103858172059
It's funny to hear Trump reading out this stuff in a complete monotone, like he's bored or sleepy. 500 billion dollar investment... meh, who cares? He does seem to like Altman though, thinks he's the biggest expert in AI. I suspect Altman has used his maxxed out corporate charisma on Trump as well: https://x.com/levie/status/1881838470429319235
Interestingly, there's another huge disparity between the two AI competitors. China enjoys complete dominance in electricity production and grid expansion. They doubled electricity output in 10 years, the US has been treading water. The US talks about a nuclear renaissance and is building 0 new reactors. China is building 29. It's a similar story in all other areas of power production, China produces roughly a Germany's worth of power production every single year. It's possible that Trump's 'drill baby drill' can change this but the US is starting from very far behind China in this part of the race.
https://world-nuclear.org/information-library/current-and-future-generation/plans-for-new-reactors-worldwide
There are considerable power issues with Stargate, more gas and coal will be needed to bring it online. Also another perspective on Trump's 'energy emergency' as being about AI: https://x.com/energybants/status/1881860142108377412
I see a conflict between two paradigms - power rich, GPU-poor and hyper-talented China vs power-poor, GPU-rich and grimly determined USA.
Political relevance? Trump perks up for a moment: "We want it to be in this country, we want it to be in this country, China is a competitor, others are competitors..." Deepseek has a wildly different vision for AI, open-sourcing everything the moment they can. I don't need to quote the ridiculously benevolent-sounding interviews where they talk about the honour of open-source for engineers as a big part of their strategy - they've backed up talk with action.
I find it quite ironic that the censored, dictatorial Chinese are the bastions of open-source technology for the everyman while the democratic, liberty-loving Americans are plotting to monopolize the most powerful technology in history and shroud it in secrecy. I suppose that's another cultural difference between the West and China. We have been fixated upon the dangers of robotics and mechanization for 60 years if not longer. We have the Daleks, the Cybermen, Skynet, HAL, GLADOS, the Matrix, SHODAN, Ultron, M3gan and so much more that I've not even heard about. Our cultural ethos is that AI is supremely political and dangerous and important and must be considered very very very carefully. There are good AIs like Cortana but even Cortana goes rogue at points.
The Chinese have... a goofy meme robot with a cannon on its crotch. As far as I can see, that is the absolute closest thing they have to a malign robot/AI in their cultural sphere. There's nothing with any broad cultural salience like Skynet, only this silly little guy: https://en.wikipedia.org/wiki/Xianxingzhe
I think this may be relevant to how AI programs are developed and the mindset of AI development. Google, Facebook and Anthropic are all spending huge amounts of time and resources to be careful, making sure everything is safe and nontoxic, making sure they're genre-aware, carefully considering the worldshaping effects of the algorithm, ensuring the AIs can't escape and betray them. Someone in Google really wanted there to be black Vikings in this world and made it so in their image model. Whereas the people at Deepseek don't really care about politics beyond holding to the party line in the required areas, they go: 'don't be such a pussy, advance forwards with longtermism'.
It actually can tell there are three R's in "strawberry". The actual log is hilarious, as it's struggling with the fact that every time it counts it finds three of them and goes, "Wait a minute, that doesn't seem right. Let me double-check the spelling of "strawberry.""
And it still can't solve this itself:
After some extensive hinting it finally gets it:
More options
Context Copy link
I am mildly impressed
I'd like to see "the infinite blink server crasher Temporalis build but it's a cautionary lullaby" now. Although I doubt it's in the dataset yet.
Some flavors of the Forbidden Build should be old enough.
In fact I just asked out of curiosity and R1 was able to figure out and explain how wardloop works with minimal nudging on my part, even just explicitly calling it "wardloop" was enough. I like how it shows the thinking process, very cute.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Schizoposting: Altman doesn’t have rizz. I won’t call it a cabal (in part because it actually isn’t), but there’s just a group of very well-connected gay men around Altman/Thiel in SV who are very powerful and constantly back each other up. Trump, in a way, enjoys the company and attention of gay men (I don’t mean in a bi/gay way himself, but in the sense that I truly believe he has a certain affinity for gay men and always has), and is kind of a straight queen, and so it all works pretty well.
Trump has always been deeply interested in image, not just his own or what's broadly anything associated with him, but the objective and subjective perception of anything. Certain gay men are better equipped than anyone else to talk to Trump about specific angles, historically certain black men and women too, but some of that was burned either by Obama skepticism and much more once he was the frontrunner in 2016.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link