site banner
PaperclipPerfector  2mo ago (text post)   33970 thread views

Culture War Roundup for the week of September 9, 2024

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.

8
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

It's late and I'm almost asleep but let me get this straight: did they just basically take 4o, wired a CoT instruction or three under the hood as an always-on system prompt, told it to hide the actual thinking from the user on top of that, and shipped it? (The reportedly longer generation times would corroborate this but I'm going off hearsay atm) Plus ever more cucked wrt haram uses again because of course?

Sorry if this is low effort but it's legitimately the vibe I'm getting, chain of thought is indeed a powerful prompt technique but I'm not convinced this is the kind of secret sauce we've been expecting. I tuned out of Altman's hype machine and mostly stan Claude at this point so I may be missing something, but this really feels like a textbook nothingburger.

Traditional CoT techniques involve prompting the model to respond with its reasoning step by step. o1 is still doing that. But what's different here is they've figured out the optimal way to chain thoughts of reasoning together. All without using complicated, inefficient RAG databases to optimize the response. Effectively, the model is trained how to "reason" by using the most effective (read predictive) strategy for CoT.

It's a coder's model I think, not a gooner's model. And I think we're only going to get the monkey models at this point. The Soviets had export versions of their tanks made with shit steel and without the good optics. They didn't want their full abilities to be exposed. Plus the export versions were cheaper, suitable for mass production.

OpenAI has the compute to train really big models, they don't have the compute to make them available to the hoi-polloi. There are already pretty prohibitive rate-limits on the new o1. And they don't want to expensively let other companies inference out and replicate their models, which may be why they're concealing the true chain of thought.

It's a coder's model I think, not a gooner's model.

I have no hopes for GPT in the latter department anyway, but my point stands, I think this is a remarkably mundane development and isn't nearly worth the glazing it gets. The things I read on basket weaving forums do not fill me with confidence either. Yes, it can solve fairly convoluted riddles, no shit - look at the fucking token count, 3k reasoning tokens for one no-context prompt (and I bet that can grow as context does, too)! Suddenly the long gen times, rate limits and cost increase make total sense, if this is what o1 does every single time.

Nothing I'm seeing so far disproves my intuition that this is literally 4o-latest but with a very autistic CoT prompt wired under the hood that makes it talk to itself in circles until it arrives at a decent answer. Don't get me wrong, this is still technically an improvement, but the means by which they arrived at it absolutely reeks of crutch coding (or crutch prompting, rather) and not any actual increase in model capabilities. I'm not surprised they have the gall to sell this (at a markup too!) but color me thoroughly unimpressed.

That's basically half of it.

The other half is using the good responses as a signal in RL on the model. An interesting comparison would be vanilla 4o with the built-in CoT techniques and the RLed model.

One interesting thing is for the hidden thoughts, it appears they turn off the user preferences, safety, etc, and they're only applied to the user-visible response. So o1 can think all kinds of evil thoughts and use it to improve reasoning and predictions, so long as they're not exposed explicitly in a way that would Harm the end user.

One interesting thing is for the hidden thoughts, it appears they turn off the user preferences, safety, etc, and they're only applied to the user-visible response.

So o1 can think all kinds of evil thoughts and use it to improve reasoning and predictions

Judging by the sharp (reported) improvement in jailbreak resistance, I don't believe this is the case. It's much more likely (and makes more sense) to make the... ugh... safety checks at every iteration of the CoT to approximate the only approach in LLM censorship abuse prevention that has somewhat reliably worked so far - a second model overseeing the first, like in DALL-E or CAI. Theoretically you can't easily gaslight a thus prompted 4o (which has been hard to jailbreak already in my experience) because if properly "nested" the CoT prompts will allow it enough introspection to notice the bullshit user is trying to load it with.

...actually, now that I spelled out my own chain of thought, the """safety""" improvements might be the real highlight here. As if GPT models weren't sterile enough already. God I hate this timeline.

This might be exactly what they've done. But... it just might work?

It reminds me of the "query hacks" for Dall-E 2 that stopped being necessary in Dall-E 3. "Picture of an elephant. Stunning, beautiful, viral on Twitter, award-winning". It makes sense that integrating query hacks into the technology directly would be more effective than leaving it to the user.

... it just might work?

It might, hell it probably will for the reasons you note (at the very least, normies who can't be arsed to write/steal proper prompts will definitely see legit major improvements), but this is not the caliber of improvement I'd expect for so much hype, especially if this turns out to be "the" strawberry.

The cynical cope take here is - with the election on the horizon, OpenAI aren't dumb enough to risk inconveniencing the hand that feeds them in any way and won't unveil the actual good shit (IF they have any in the pipeline), but the vital hype must be kept alive until then, so serving nothingburgers meanwhile seems like a workable strategy.

Something like those hacks is probably still going on behind the scenes thanks to insertions being made by ChatGPT when you use DALLE. The reason those hacks work is they help the model to zone in on the desirable part of the latent space, and DALLE3 is still a diffusion model, so there’s no reason to expect it works fundamentally differently from other diffusion models.

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