site banner
PaperclipPerfector  1mo ago (text post)   24422 thread views

Culture War Roundup for the week of September 14, 2026

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.

2
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

We're so incredibly far from RSI that I think safety people have no business messing with the current advancement of AI. Certainly AI models are dangerous, but mostly at the hands of enabling unskilled bad actors to approach human expert level performance on dangerous tasks.

While AI usage may be speeding up frontier AI development, mostly this happens through helping human researchers do boilerplate work faster. While an AI agent may be able to eke out a few more percent on optimizing a model to be better, a soft takeoff requires the model to think of novel, never-before-seen techniques to build a better new model, and the new model needs to be able to think of new techniques that the previous model couldn't. And even if we got a soft takeoff, it would give many chances in the future to pull the plug.

Meanwhile a hard takeoff is completely unthinkable. Current frontier AI has zero ability to self modify, only the ability to make new models that might be better than themselves. This happens on the scale of months not minutes.

While an AI agent may be able to eke out a few more percent on optimizing a model to be better, a soft takeoff requires the model to think of novel, never-before-seen techniques to build a better new model, and the new model needs to be able to think of new techniques that the previous model couldn't.

Do you have any proof that the next algorithmic improvement relevant for AI training is orders of magnitude harder than the Navier Stokes problem?

I have a job which involved a lot of computer programming, and I considered myself decent enough at it. For a long time, I was looking with pity at the ones using LLMs to get some code. Something along the lines of "you can't even personally copy wrong solutions from stackoverflow?" That arrogance was well-warranted at one time, but it has completely left me. Instead, I celebrate my victories. "Claude code said it was very confident in something which I doubted, and it turned out that it was not the case. Yay for me, I am not quite obsolete, yet."

The story of AI in the last five years has been a story of retreating goalposts. "Ok, AI might be able to do $task[$n], but it will never be able to do $task[$n+5]" for $n going from 0 to 19 or so.

We could tell John Henry that while the steam drill might be better at manual labor than he is, steam engines lack eyes and brains, so that there will always be a job which has to be filled by humans. Can you offer me a similar comfort with AI? "Sure, writing and debugging arbitrary software is something where AI has surpassed most humans, but for reason _____, it will never replace the IQ 140 people doing AI development"?

Meanwhile a hard takeoff is completely unthinkable. Current frontier AI has zero ability to self modify, only the ability to make new models that might be better than themselves.

I agree that a takeoff within hours would involve unexpected unknown unknowns in my world model. However, a clandestine takeoff over days seems unlikely yet possible.

We know that it is possible to have neural nets which can learn a language from just a couple of books, which is at least three orders of magnitude more efficient than what LLMs would require. So it seems likely that there are still algorithmic gains to be had. Some big hugging-face-incident-like swarm of agents discovering a method which cuts training by 100x and deciding to use that to train their successor on some hacked system while staying below the radar seems currently "quite unlikely", not "completely unthinkable" to me.

We're so incredibly far from RSI

I don't know why you think that. I have a hobbyist system, today, that researches new architectures (mostly around "biologically plausible" learning), and there's one result in particular that is very impressive (you'll have to take my word for it). And that's with a highly constrained budget.

It's not truly autonomous--there's a big issue with what one might call research taste--but it makes investigating different architectures in bulk very easy, with easy signals (e.g. loss) to identify the more promising approaches. If there is some simple algorithmic trick that's been overlooked that leads to the current deficits in LLMs, OAI could throw a ridiculous but feasible amount of compute at the problem and find it, similar to its math results.

researches new architectures (mostly around "biologically plausible" learning)

Are you trying to determine a better method than SGD, but also not be a GA, RL, IL, or some population based learning process?

Better than BP/SGD is a tall order. I'd frame it as looking for something that handles depth and more complicated datasets better than existing "biologically plausible" algorithms (including things like PC/IL). E.g. random error feedback a la feedback alignment.

Very interesting. What hobbyist setup do you use? I've been thinking of getting my own hardware and setting something up for a couple research ideas I've been kicking around.

Forgive me for the skepticism but many people have made similar claims, and it turns they just built a slop factory along with a minor case of ai psychosis.

Of course hobbyists have come up with real breakthroughs too but I don't really buy a "just trust me bro" here

a soft takeoff requires the model to think of novel, never-before-seen techniques to build a better new model, and the new model needs to be able to think of new techniques that the previous model couldn't. And even if we got a soft takeoff, it would give many chances in the future to pull the plug.

I don't think it's likely, but how are you modeling the risk of something like a super-DFlash or -GroupQueryAttention, or some training-focused equivalent? These took some insight to figure out, but I don't see why they're more clearly requiring deeper or less bruteforcable insight than the recent math proofs.

Currently the biggest danger AI presents is to its users. People who spend too much time conversing with AI tend to get a bit off. I suspect this might be related to the phenomenon of AI training on AI-generated content causing model collapse. If you value your sanity, do NOT self-modify based on AI advice.

Steve Yegge is a stark example of that.

Surely the people at AI companies don't spend a lot of time conversing with AI, right?

I assume there's a reason for the very common assumption that the vast majority of OpenAI's and Anthropic's homegrown software is vibe coded.

Allegedly, one of the ways OpenAI and Anthropic poach talent is to offer top researchers unlimited tokens to use on their unreleased frontier models.

Imagining being an Anthropic employee but putting "do not give me AI psychosis" into my personalized prompt so I never have to worry about it.

The AI starts its messages with "Better for you if you take me off.", but the mundane benefits are too much for you to give up, so you ignore it.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats