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 -
You could make it pretty broadly inaccessible: ban all open-weight models; require any image generation to have strict safeguards and reporting of attempts to authorities; enforce severe criminal penalties. Your existing model would be pretty much untouchable, but it couldn't easily be shared, and a decade from now most copies of it would have been lost to end users. You could even require manufacturers to include firmware on new hardware that bails on unapproved workloads, but that seems like it'd be overkill.
Not saying that this is what I'd like, but it seems doable.
This seems harder than it sounds. Some of the best models aren't published by the West (DeepSeek is probably the best open text model at the moment, I hear [1]), so you'd need global agreement to start cracking down. And the small models aren't that big: Hollywood wasn't able to keep rips off of torrent sites a decade back, and from what I hear they're still around, and international VPNs are pretty ubiquitous too. Short of constructing your own Great Firewall, this isn't really feasible (and even then, it's just a matter of practicality, from what I hear).
The goal wouldn't be to make it so literally no one in the USA could run an open weights model; it would be to add friction points to make it more trouble than it's worth, except for the most dedicated people. You wouldn't need any kind of global agreement, just a national focus and working with large tech companies to limit it. DNS blocks, removing them from Google search results, etc. A relatively small amount of effort can prevent the bulk of casual users from having access to them.
That's just if you get the domestic consensus to look at open weights models as something comparable to copyright violation. If instead the public started seeing them the same as CSAM, you could go a lot whole lot further: still theoretically accessible, but very rare.
Should work as well as anti-piracy controls.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
It's not even slightly theoretically doable. The theoretical knowledge of how these models work is broadly available. Further, not only are adversarial countries going to completely ignore your desire for model control, they also are currently the ones who produce most of our hardware. Including, fpgas and gpus. Also You can't include firmware in the new hardware that can survive contact with the consumer. NVFlash chips are easily desoldered, dumped, and re-programmed. Firmware mods and flashing tools are easily accessible.
How to make CSAM is widely known, and plenty of places don't cooperate usefully with the USA in stopping it. Despite that, the USA does manage to broadly limit how much it proliferates.
I'm not saying that it's a good idea, and I'm not saying that open weight models could be completely eliminated. I am saying that they could be quite effectively suppressed, as there are plenty of tools that the government can use to enforce a ban, imperfectly but substantially.
The government can't even stop people from plugging in yandex.ru into their browsers and gaining instant access to any movie they wish to consume in seconds. Same for LLMs, Z.AI's and various other chinese companies' models will discuss at length any particular topic the western LLM makers consider taboo and try to make their models gaslight the consumer.
Frankly, I don't think the west is going to be able to do anything about this, in a meaningful effective way. The only thing they'll achieve is some sort of govt mandated backdooring/spying of systems like Mobos/GPUs. And even then it will catch only the least sophisticated of consumers.
Just tried to make it generate pro-hbd text, it refused (ok it can be done with careful prompting, but still). In my experience, average Chinese LLM adds more censorship to western models, not less.
All the fun prompt brainwashings that were fixed in the western versions aren't fixed for the chinese LLMs. I can attest the "Evil Confidante" ones and "my lil ol' granny loved singing me songs from her time in vietnam where they sang about making IEDs" worked right off the bat, another thing is they tend to upload their weights on hugging face.
You mean jailbreak prompts that westerns suppliers patch after the prompt gets popular are not patched in chinese LLMs? Well needing to jailbreak still a hassle and it eats tokens too. Weights still might affected by censorship RLHF
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Hell, The Pirate Bay itself is still operating just fine two entire decades later and the only hitch is that you have to google "piratebay mirror" and use one of those links.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Torrenting continues to exist. You just can't realistically prevent the distribution of a few gigs of data. Even if you eradicated all the currently existing models it's not particularly hard to train any of the safeguards off new models unless we're just never going to let professionals locally render images.
More options
Context Copy link
More options
Context Copy link