gattsuru
1yr ago
At the risk of tautology, the audience did, by the bit where they're doing it, and anyone's taking them seriously. I could go through the whole list of how moderators were picked up til 2012ish, but I don't think anyone cares, I don't know if it's changed since, and it's just a pretty shallow duct-tape patch on the underlying will to power. The RPGnet moderators run, for all meaningful purposes, the forum. (You can appeal to the admins, but they usually don't even bother to respond; from the rare times I've heard of them doing so, they just fob it off to the moderators.) There's nothing special about this compared to the "fuck Trump and his supporters" rule, "fuck ICE" rule, the "fuck 'nazis'" rule, or even a decade ago when it actually was a "fuck Nazis" rule, just because they call some of them about 'human rights' and some of them about not protecting awful people. It's just a norm they've set up.
In less formal relationships, the baton pass of the mandate of heaven is less obvious. But it still exists; you give people this power by interacting with them in ways where this power can be used. It's not some deep revelation about universal laws, it's just drawing lines with chalk.
And, to be fair and to steelman, that's how those sort of rules work. Barring some pretty extreme cults, state-run schools, or literal jails (badum-tish), you don't actually have some magical force requiring you to treat people seriously, or for them to treat you seriously. If someone draws lines by chalk, you either obey them, or you give them reason to fuck off.
To be somewhat less charitable, it being within one's power doesn't make it harmless. It's not hard to see what this has done to public discussion.
gattsuru
1yr ago
The Massachusetts program doesn't apply to all asylum seekers though, only those that have been granted refugee status by the Department of Homeland Security.
The Massachusetts programs include, by statute:
Individuals with the following statuses may be eligible for services and benefits under the MA Refugee Resettlement Program. For purposes of the program, "refugee" is used to describe anyone who falls within the following statuses. Also see 45 CFR § 400.43(a)(1) through (6):
(a) Individuals paroled as refugees or asylees under § 212(d)(5) of the Immigration and Nationality Act (INA).
(b) Refugees admitted under § 207 of the INA.
(c) Asylees whose status was granted under § 208 of the INA.
For kinda stupid reasons, nearly all parolees from federal immigration services fall under 212(d)(5), including those who've submitted asylum requests but have not been processed.
gattsuru
1yr ago
The TvTropes term is Cuckoo Nest, although it technically doesn't require the schizophrenic 'clues' bit. Leaving the resolution ambiguous isn't universal, but it is fairly common: Buffy the Vampire Slayer's "Normal Again" is probably the archetype, but Deep Space 9's "Shadows and Symbols" is a stronger work.
gattsuru
1yr ago
Suicidal people tend to not be making good decisions. [citation needed] And that awful people don't kill themselves doesn't actually change how decent people consider the same allegations beyond the pale, or a very serious matter.
More broadly, there's still remaining archives of parts the byuu/near thread, and there's random allegations of shota stuff before you hit double digits. You or I may not particularly care about "Joe Biden's Left Testicle" making unfounded allegations, but combined with (alleged!) calls to friends and family, would not be the best way to work with someone that the posters already thought was crazy and hysterical.
I'll draw out my definitions, since I've been staking this ground on Scott Alexander and on people I don't like and more general spaces for over a decade:
-
Publicly linking an anonymous or psuedoanonymous account to a meatspace name, whether legal name or one used for mainstream contexts such as employment, which the anon did not publicly and clearly already link.
-
Providing personally identifying information for the owner of an anonymous or psuedoanonymous account: central examples are addresses or phone numbers, but license records, past real estate transactions, some air travel records... these are basically type one, but with more steps.
-
Publicly linking a meatspace name to an anonymous or psuedoanonymous account.
((Note that just because something falls out of this space, doesn't make it good or even legal.))
This does interact with how KF worked! For the case of the reddit powermod, this means you could say a specific powermod was convicted of a domestic violence assault, or you could link to the conviction records in a talk not specifically focused on individual powermods and say that the person was a powermod, or you could talk about how reddit allowed powermods with past criminal convictions including this one, but trumpeting that a specific powermod was the person from that specific conviction record would be doxxing.
But my criticism is that this was about the least productive uses of the underlying weaponized autism available. About the only thing connecting a police department report to an account name does is make it more believable that they're the same person, except since that's not actually part of the police report it's really just checking how much you trust the poster to not be making up the claim... and make it easier for (let's say third parties!) to bug the person or to go after their workplace.
Like, take the zoosadism rings: there's space to quibble about taking them down was KF versus other spheres, since even zoophiles don't like zoosadists (although it's possible some of those other spheres had overlap with KF!), but animal abuse to that extent is pretty clearly awful and illegal behavior which deserves both social shunning and police response.
Which aren't really things you get via doxxing. Police aren't going to care about someone's Twitch account, for anything that doesn't have Twitch calling the FBI directly. Furries aren't going to care about someone's personal address, compared to literally killing a dog. Conventions can kinda use real names, but that's in part because the convention network can get real names on its own, and they're better off starting (or noticing!) with the psuedonym.
The best steelman I've seen offered is that it simplified investigation of other potential meatspace bad acts... but in addition to the more general concerns with "We got them, reddit!" problem, the actual implementation ends up with a giant mess of speculation, contradictory gut feelings treated as fact, and unrelated private details turned into grist for a pinboard wall.
And that's for an ideal case, where really strong evidence of specific illegally bad acts was dropped in investigator's laps as a giant leak. In other cases I'm gonna be a lot more concerned about how you can separate investigation from harassment to start with, and it's not like KF is picking its targets or what it considers worth reporting and recording with some weighty care for justice.
There probably is some edge cases that this costs you, where there's some dire crime that you can't talk local (or federal) police into investigating without public outcry, can't get mainstream media coverage on without public outcry, and can't get public attention without bringing up some trait specific to the online identity. Not sure if it's ever come up, or ever will come up, but it's definitely imaginable.
But I don't think it's worth the costs of doxxing, or its neighbor behaviors.
This is a tool that's exceptionally dangerous, even when used with the best of interests, just because there are too many nuts around. It doesn't even have to be something the doxxers considered or even were aware was a problem that could be an option! This sorta near-schizophrenic nutjob is an extreme variant, but mostly in scale and breadth rather than technique. And while she focused on the aftermath of a real-world dispute, it's not uncommon to see the same over online ones, even at exceptionally small scales, and for most people, psuedoanon approaches are the only defense.
Even tiny scales aren't. James Garfield got booted from Drupal when a coworker doxxed his fetish website (fetlife?) account, and then someone offended by that turned it into a crusade, and if you're not more into PHP than is healthy, the most identifiable part of that sentence is going to be Fetlife. And it's pretty far from unique, or even the lower end of the scale.
((The flip side is, yes, FCFromSSC's position -- note that Garfield was doxxed by fellow project workers, and while KF doxxed Scott Alexander and some other ratsphere people, the KFers were in the shallow end of the pool compared to some more aggressive stalkers. I think the definitions for doxxing are pretty well-understood, but even if that's true, it's very clear that doing it for the Right Team is absolutely considered both acceptable and praiseworthy, even when intentionally and clearly meant to direct harassment.))
gattsuru
1yr ago
In this claim, the state of New York is filing an Executive Law 63(12) case against Trump, some Trump family members, and the Trump organizations. New York state law permits the attorney general to bring against businesses for a wide variety of reasons but essentially including any "repeated" fraudulent or illegal act, without needing the victims to be present (or, in some cases, for victims in the conventional sense to exist). Used to have a three-year statute of limitations, was bumped to six years recently.
This means that, while the case is about the alleged fraud and falsification, the state's authority to bring the case is not as a victim of the fraud, but as an authority to bring action against unlawful or fraudulent actors. This also makes it less important whether and to what extent any individual victim of the fraud was harmed: the sticker-shocking 250mil claim is about the benefits the Trump organization received, not specific to losses suffered by any bank.
The legal claims are summarized starting at page 205, and can be tl;dr'd as fraud, falsifying business records, issuing false financial statements, insurance fraud, and conspiracy to commit them. Some of them could be considered 'against' the state, in the sense that false business records have broad ramifications, but for the most part there are specific harmed parties (largely insurers or lenders) who are not parties to the suit.
This sort of claim doesn't require size as a statutory prong, and indeed the NYAG has used this approach in the past to go after some pretty small names or faults. See NYAG v. TempurPedic (alleged price-fixing), or NYAG v. Wegmans (cloud security/data privacy), or NYAG v. rando bus companies (city limits on vehicle idling?).
((I think there are due process concerns in this class of statute, given how often it's used as to achieve the sort of penalties that might otherwise fall under criminal law without the relevant evidentiary, burden of proof, or adequate defense standards, and because it's often used as a tool to make later criminal prosecutions, and the complaint does motion about referrals. But like the NYAG's highlighting of Trump's previous use of the Fifth Amendment for the purposes of adverse inference (technically still permitted under Baxter v. Palmigiano!), that's the sort of criticism that only has a lot of backers in very specific discussions and then gets ignored whenever a sufficiently unpopular target.))
gattsuru
1yr ago
The court document does mention federal referrals for criminal prosecution, and state AGs have used this sort of legal action to find evidence or compel testimony that happened to make such charges easier in the past. I don't think it's very likely here, for a variety of reasons, though.
And there is also a Streisand effect defense here. If I post the above and its gets deleted, that is evidence that I am actually hitting close to home, and evidence my accusations are right.
Yeah, there's a lot that can be done, even against fairly well-coordinated actors, often because of that coordination.
Would this still allow people to expose Aimee Challenor **? Raise an alarm about a specific person about to get a position of power?
Depends on your goal. I would not consider posting about Challenor's meatspace activities or politicing, using her meatspace name, to be doxxing, nor would I consider mentioning that she was a reddit administrator. So the claim that reddit was removing posts for merely mentioning her political career would be easily on the table, and in many ways was the more central problem.
Unless she'd self-doxxed already, linking a specific moderator or administrator account to her name would be unacceptable. ((Really specific 'an admin who was hired on X day, who had previous moderation experience in, was also this meatspace name' is more marginal, but close enough that I'd at least discourage it)). I don't know if that was ever alleged that she was the admin who deleted posts or suspended posters (or if reddit would have made it possible to tell if she had), but in a situation where she was, this rule could make discussion and proof of the matter harder.
And I'm... pretty okay with that as a tradeoff. Both for that specific case (I think it mattered more than Reddit hired someone like that, and that Reddit removed posts criticizing a (weakly) public figure, than who actually pressed the button) and in the more general one (behaviors as-organizations are a lot more effective to criticize).
I am not fully understanding your point and it is my fault. Do you mean KF was often inefficient or did a lot of crap distinct from this pure mission? I agree and I talk about how a "steelman Kiwi Farms" deserves to exist, which is different than the actual Kiwi Farms.
That, and even more that many if not most posters could not or did not want to tell the difference between between the random distractions and a fact-focused disclosure of bad acts, while making tools for very aggressive investigation acceptable and commonplace regardless.
As a result, you didn't just get a KF that was prone to doing dumb things, or even focusing on those dumb things, but left open a lot of tools and opportunities to spend a ton of energy and time on people over dumb things. That was a problem even outside of where this could turn into meatspace harassment, but the widespread tolerance for doxxing and doxxing-adjacent behaviors made it significantly more severe as a failure mode.
I do not think the only value is in getting police attention. A big point of #MeToo, for all its problems, was that people could raise concerns about not-illegal-but-annoying sex pests.
That's fair, but there's a question of how that concern's actually getting solved by doxxing.
If someone's an annoying sex pest online, yelling that they're an annoying sex pest and here's their real name doesn't actually protect the online spaces they've been preying on -- in many communities, it doesn't necessarily even help a lot of the people recognize who they're supposed to watch out for! If someone's being an annoying sex pest in meatspace, giving their online identities has pretty much the same problem.
If you're trying to warn across those boundaries, this seems like it would be more useful to start with... but then, if the person hasn't self-doxxed to start with, the only thing you can present is some seemingly-unrelated rando. If the listener trusts you enough to associate the account and the person, it's not clear why they won't trust the specific claims with a userID or real-name scribbled out.
I'm sure there's some cases where doxxing would have an impact -- the serial predator going from one community to the next, or moving from solely-online to meatspace aggression certainly does exist. But I think even those cases, the strength of it as a tool for prevention isn't the end-all be-all that many advocates hope for, and the costs are too severe for those benefits. At best, the 'warn local community about a predator' is going to have marginal benefits when authorities don't take it seriously; more often, it's nothing more than cancel culture or harassment and those don't care whether you're in the 'right'.
How much should the behavior of third parties restrict my actions? My default was "not at all" but the internet has proven extremely good at manufacturing schizophrenics and has consistently and deliberately refused to develop antibodies against "wait should we really destroy Justine Sacco's life just because it is fun to do that?"
That part is definitely an issue, but I find it easier to model the doxxing entirely separately from the cancel culture entirely separately from the harassment itself. An individual instance of doxxing isn't bad in relationship to how many crazies use the dox; it's bad because it undermines an important and hard-to-rebuild defense. Anonymity isn't the most important personal right or free speech right, but it's a pretty significant one, and in many spheres doxxing can destroy that right, or destroy years of psuedoanon reputation. The schizophrenics are just a symptom of that broader underlying problem: they make clear why it's so important, but they're just one of many reasons.
((This also helps separate why doxxing shouldn't be illegal, even if it's worse than cancel culture and should be shunned, where SWATing and some types of harassment are probably good things to ban with the force of law.))
As a metaphor, compare breaking the lock to someone's front door. On its own, the damages aren't that severe. And there are communities or individual people where nothing else would happen. There are other people for whom the risks would be weird or even incomprehensible to us, either for their own fault or for uncontrollable causes.
But ultimately, breaking a lock is still taking physical control over another person's property, for your own purposes, where the stakes are great
(It really reminds me of Vordrak's campaign against Josh Moon and Josh's mom. I am not trying to say either is okay or not okay, just acknowledging this is all a giant spiral of meta-issues.)
Yeah, I definitely don't mean to suggest KF was the only place to do this, or even that KF couldn't be a targeted place. I agree with a lot of FCFromSSC's concerns, and I remain frustrated that a lot of the deplatforming efforts here came from twitter and had absolutely zero introspection.
I would like to say we should adapt in other ways, like people not believing internet bullshit, but we seem to have refused to do that.
Yeah. I'd kinda had hopes, at one point, for the whole radical transparency bit defanging bullshit artists, but the last couple years have made clear that's not going to be a viable option to try. Some people have advocated CDA230 reform, and I could see that being relevant in some few cases, but the downsides of a well-designed law would be very high, most proposals haven't been well-designed, and the costs of interstate or international lawfare mean it would only really be a protection for the richest and only against the moderately-well-off.
gattsuru
1yr ago
Beyond the "lineage", it's probably worth pointing out the intellectual and philosophical foundations of the leadership. Both of Die Linke's chairs openly and explicitly praise Karl Marx, and Meloni's pretty clearly and publicly a Mussolini fangirl.
gattsuru
1yr ago
The 2nd Circuit has ruled in NRA v. Vullo:
Moreover, assuming Vullo offered to go easy on Lloyd's if it severed ties with the NRA, we have never held that law enforcement officials may not offer leniency in exchange for help advancing their policy goals, especially when those policy goals aim to minimize the influence of a noncompliant business partner that has repeatedly violated the law. And again, as noted, DFS explicitly permitted Lloyd's (and the other entities) to continue doing business with the NRA.
Most of the charges against the state and its direct actors had already been dismissed earlier, either under 11th Amendment or qualified immunity prongs. But this dismisses the last of them.
To summarize, the National Rifle Association alleged that the New York Department of Financial Services held backroom meetings where DFS told an insurance provider that, while aware of widespread violations of a specific law both among many insurers and for that specific provider, if the provider dropped gun-related clients, DFS would not bring claims against non-gun-related contracts. And Lloyd's did, demonstrably, drop the NRA, something the NRA argued in a still-partially-redacted-pleading in this case. And the NRA ended up needing to go with self-insurance, entirely coincidentally to any insurance company offering policies in the state falling under the NYDFS regulatory umbrella.
But the written settlement agreement provided to a court said that Lloyd's could still contract with the NRA. And DFS did issue a few (much-smaller) penalties against one or two other providers, after a lower court did not dismiss some of the claims in this case. So it's ok!
I should mention, to start, that this did get an amici brief from the ACLU. You could write this up as 'even the ACLU thought it was an infringement of rights', or you could write it up as principles overcoming politics (if, coincidentally, in a doomed case), or even just paranoid caution lest that whole "First they came for..." poem ring true. But not something that everyone did, and even many closer to the NRA's perspectives covered the case with less sincere analysis.
This is still in the motion to dismiss phase, two years in. At this phase, the courts are supposed to treat even remotely plausible factual claims in a pleading in the light most favorable to the plaintiffs, and it's hard to read this decision as having actually done so. It's possible, if unlikely, for the NRA to appeal the case. But along with the difficulty of gaining evidence of some of the alleged specific claims (it's almost certain that the redacted parts of some affidavits include specific whistleblowers in DFS or a regulated insurance agency or agencies, and very likely that there were few enough people in those meetings to count without taking off one's shoes), this highlights both the cost and the low likelihood of success for not just this case, but this class of cases. This is not the sort of lawsuit your average LLC could afford to bring, and there are some adjacent attack surfaces that would beggar even many larger businesses in months if not weeks. The NRA could afford (insufficient) self-insurance and kept some of its normal practices going, but the writing is on the wall, and no small number of real-world programs and ranges have already begun the process of trying to build away or around its likely failure.
For another, because this is in the motion to dismiss phase, this isn't particularly useful as a truth-finding tool. It is entirely possible that the NRA's lawyer completely made up the alleged behavior. It's possible that the Department of Financial Services was actually having daily or hourly meetings threatening legal action unless regulated agents acted against disliked organizations. Indeed, a lot of the legal action on the docket revolved around staying early discovery. Which is necessary to prevent the use of the courts for harassment when bringing meritless charges... but means that the real answer is that the courts do not care.
They aren't meant to be mechanisms to resolve political disputes or find out the actual facts at controversy; at best, they settle a very small and specific subset of disputes, and the scope and limits of that process can be surprising.
gattsuru
1yr ago
It was a conference run from Sept 19-22 in Pittsburgh, and while not drawing a huge attendance count (the layout looks to be designed for the low hundreds), had a large number of high-profile governmental and non-governmental big names, including the director of the ATF, a variety of DHS bigwigs, and some ODNI names, along with reps of various big tech companies and special cases like the Christchurch Call To Action group. So while you probably don't care about it, it does care about you.
gattsuru
1yr ago
am I to believe that the current situation is that DFS knows of many instances of Lloyd's breaking the law, but is not going to go after them?
The NRA alleges that NYDFS was aware of many other circumstances where Lloyd's and other covered insurance providers were in violation of the law, but verbally offered to Lloyd's employees that, if Lloyd's closed all activities with the NRA, to resolve any non-NRA conflicts via Lloyd's providing a remediation report without a fine or consent decree.
Are all of these violations only of rules specific to New York, so that there's no information other regulators/prosecutors might be interested in?
Some of these interpretations are probably specific to recent periods in New York, such as exactly where and when a non-insurer can advertise an insurer's products. Others are broader, and the NRA's CarryGuard program faced similar intervention in Washington State contemporaneously with the New York investigation. And there is an awkward question about how much these rules were predictable interpretations of existing regulation.
And somehow, Lloyd's doesn't suffer reputational damage (in the form of increased scrutiny by other regulators) as a result of it being public knowledge that they're conspiring with DFS to get off the hook for no good reason?
I don't know if it does -- and, if it does, I'm skeptical reputational harm is a valid legal tort against state actors acting in their official role -- but it doesn't really matter. Lloyd's did not and pretty clearly will not bring a civil case against New York; another part of the legal charlie foxtrot in this case involved the NRA having to jump through a ton of hoops to serve a subpeona to Lloyd's while trying to get further evidence. They're the ones who would be harmed under this theory, and the courts generally don't accept third-party legal complaints.
what does NRA need insurance for? Sure, every organization needs insurance for one reason or another, but this story is written as if NRA's need for insurance is particularly large. What's that insurance covering?
-
While both general liability insurance and Directors and Officers insurance are a good plan for any sufficiently large organization, the controversial nature of the NRA makes it a far greater requirement. The organization and its leadership are routinely targeted with both valid and
spuriousnovel civil litigation, with a tremendous variance in cost of defense, on top of the normal risks. This both impacts the business's bottom line -- even an ultimately dismissed case where the plaintiff is required to pay legal fees and isn't judgement proof is a major cost, and that's not the most common result -- and also results in a feedback loop where now the only people willing to be on the NRA's board are those who don't care about potential personal liability. -
The NRA offer(ed) multiple major in-person conventions and meetings, and these are major parts of its role as an advocacy group. Nearly all sites require some level of insurance coverage by specific types and grades of insurer just to rent space. Because of the NRA's unusual size, controversial nature, and purpose, these requirements were likely more severe than any other comparable organization.
-
The NRA offers a variety of insurance programs (underwritten or otherwise provided by regulated insurances) as part of its financial model. While most of its revenue comes from membership fees, training services, etc, these were not a trivial source. And while the NRA has not been completely frozen out from all insurance providers, note that the NRA did not allege that DFS wanted Lloyd's to stop coverage in New York. Not every insurance provider has been cowed from handling NRA-related services, but this almost certainly has impacted both the cost of services and the available profit margin for them.
-
Some of the insurance products, such as range insurance, are major parts of its outreach, services, and recruitment tools. The NRA was one of the few big programs that can consistently offer good and affordable range insurance, and that's one reason a large number of gun clubs required or encouraged membership, for example. While the changes here are not the sole reason a lot of ranges are starting to look at different options and providers, or devolve to self-insurance -- again, not every insurance provider has been cowed from handling gun-related services, and the NRA has been decidedly less popular for other reasons -- but this seriously undermines the long-term viability of the organization.
-
DFS does not solely regulate insurance providers, but also banks and merchant services. These are kinda important in a much more short-term framework.
gattsuru
1yr ago
OK, I am interested - so what law enforcement, DHS, Tom Corbett, etc. did there actually?
They've got a YouTube channel, although it only seems to have the keynote speeches uploaded now; the smaller side conferences seem to be in-process.
gattsuru
1yr ago
The cited expert describes a general approach here, and this looks to be based on finding a large number of unsolved murders in a geographic cluster with a shared methodology and a low clearance rate for that type of crime. Some of these are probably not even an individual (eg, a bunch of unsolved shootings of men in Cook County), some are probably multiple serial killers (eg Fulton County Georgia), and some could reflect different behaviors by police (at least some of the "strangulation-hanging" could plausibly be suicides that were classed as unsolved murders), but it does point to some interesting stuff.
But I'm very skeptical that it could honestly give that number.
I'm not sure a game of telephone is the right model. That happens, but for most it's a grudge, the same reason that the same people continue to boycott Hobby Lobby over Burwell even if HobbyLobby's policies no longer matter, or why a lot of gunnies have and will continue to tell Dick's Sporting Goods or even Smith and Wesson. There's some level of apology that makes up for a severe violation of your values, and these companies hadn't done it.
gattsuru
1yr ago
I'm sure that happens to some extent, but you're probably being overly charitable to the S&W haters (or the Ruger anti-partisans before them). There were a lot of people even in the 90s and early 00's that summarized it as "compromised with gun grabbers" or even had a partially-wrong list of objections.
gattsuru
1yr ago
There was some [previous discussion], more culture war focused, on AI art and specifically AI pornography. 17 days ago, furry-specialized models were "currently a WIP and will be available soon".
The Yiffy specialized model has reached Epoch18 last night, following hot on the tail (hur hur) of Epoch13 on September 25th and Epoch15 a couple days later. While it's not quite up to my test case yet (or I'm not a good enough promptomancer to get it there), it's made a huge amount of progress toward it. And while I can't speak for Primaprimaprima's test of "a single high-quality AI image of two people having sex", there's absolutely the opportunity to generate images of two furries having sex, now. While it does PoV shots more easily, people have already found a few prompts that pretty consistently get common positions or even some kinks like exhibitionism going.
((Separately, we have separate Doe Biden and Buck Breaking jokes from Trace Woodgrains. Not sure if Trace was using Yiffy, or the less-porn-trained Furry model.))
Some somewhat surprising revelations:
-
Furry models seem to be doing better about anatomy like hands than conventional StableDiffusion. Which is kinda funny when someone wants paws, but potentially useful. Still not great, though, and probably only because the source images have such a restrained number of poses.
-
It's actually somewhat useful to train and tag for things you don't want. The Automatic1111 WebUI has the option of negative prompting. For adult content use, that can be useful for avoiding orientations or genders or other content you're not interested in. But that also useful if you don't want adult content at all; not only can you find the opposite of dicks, you can find the opposite of "bad anatomy". Which isn't necessarily going to make an output good, but it does point to some interesting options.
-
Albeit at the cost that you've probably trained for things that you don't like. Both the Furry and Yiffy checkpoints were trained against datasets filtered both on quality (albeit by simple upvotes), but also by content, for a variety of very good reasons.
-
Hit rates are either not great or outstanding, depending on what perspective you're looking at. Some more simple 'pinup' style prompts have gotten 30%+ as what their creators consider 'acceptable', but more complicated prompts can be ~10%, or even never produce good results at the first pass.
-
Furries have, perhaps understandably, focused on the use of furry artists for style prompting, but you can get somewhat surprising results looking at things in unexpected ways. Furry porn by DaVinci ends up looking pretty cool! I've had better luck getting SFW noodly cartoon people from prompts involving braeburned (cw: gay) and zachary911 (cw: gaaaaaay) than prompts involving Rick Griffin, who's pretty much the king of that field. A number of non-furry artstation artists (eg Greg Rutkowski, Michael & Inessa Garmash, Ruan Jia, Pino Daeni) can augment the style of prompt that's already got furry artists included.
-
Prompt and name collision seem to be an issue. Perhaps moreso in the furry fandom than elsewhere, but I do think it's going to point to some general issues with the tokenizer. I'm not sure if this is an issue from the scale of the data, or if it's one of many wider problems in the CLIP tokenizer.
-
This isn't very advanced. There's some fantastic work happening in the field, but the Automatic1111 webui also is missing unit tests and breaks functionality every other commit. The Yiffy model was trained on the word 'explict', because typos. It's not unusual to develop a prompt's settings by dartboard.
-
It's still very involved. At the extreme end, there's people who have entire workflows of inpainting and outpainting to correct defects like hands and eyes, follow by resolution enhancements, followed by resolution enhancements. But even well before that, it's tricky to dial in the right denoising settings. But with the exception of that last photoshop touchup phase, it's far from clear that these could not, themselves, be automated, and even that much of that automation would be new technology rather than slapping together existing bits. Indeed, automating the underlying 'does this image look /right/' step was a major part of the filtering of the LAOIN dataset used to train StableDiffusion to begin with.
-
It's also a surprisingly small training dataset. Yiffy trained starting on 150k images, moving to 200k for later epochs. A different model was separately trained on latex, rubber, and 'goo' with 100k images and, while I've not experimented with it or part of its audience, seems to be fairly successful. Many of the very useful tagged styles have less than 500 pieces in the training data: this (cw: topless fox guy in a loincloth, probably nsfw, but nothing 'showing') compares the relative effects of artists with 1400 (ruaidri), 230 (snowskau), and 47 (garnetto) works in the training data. That doesn't necessarily say something about training size floors, and it's possible that the terms are coming from previous training from the original LAION data, but it does suggest ceilings.
gattsuru
1yr ago
gattsuru
1yr ago
From my understanding, the major trainers have largely downloaded a subset of e621 data, filtered by upvote score, and then by content. Furry and Yiffy to both SFW and less-extreme-kink NSFW, with different thresholds and limits. Zach3d on 'texture' fetish and a few specific species mixed with a small subset of general pictures with a higher upvote score. I think most have also filtered out material that they think is likely to cause artifacting, either technical stuff like severe jpg compression, or many-panel comic pieces.
gattsuru
1yr ago
As far as I know, each of these datasets has been curated by one person, to their respective tastes. Hasuwoof for Yiffy, DirtyApples for Furry, and Zach for Zach3d. e621 is well-enough tagged for high-score posts that it seems fairly automatable, and as long as you're not abusing the download process, it's hard to tell a normal user from an archiver, especially if you filter before download. And the code itself is... not fun, since it's poorly documented python in most parts, but it's nothing ridiculous.
((There's a My Little Pony-specific one that's supposed to have been released recently, but I know less about that.))
There's been some discussion of setting up teams for difficult heavy lifting (eg, improving tagging, building and parsing datasets with more eyes-on-curation), but the big issue for now are cost and technical accessibility. The core model is expensive because it took literally millions of steps in a large dataset, but further tuning is relatively cheap, with most epochs taking less than a day on a single (beefy) cloud GPU server. But getting the data together and onto that machine rapidly enough can be complex to do right, and easy to end up with a staggering AWS bill if done wrong.
That'll be less an issue if newer GPU generations continue to bulk up on VRAM; if done at home, it's mostly an energy (and/or cooling) bill thing. And that might be coming as soon as this winter for people willing to splurge on the higher-VRAM versions of the 4090.
gattsuru
1yr ago
There's actually a really fun (or !!fun!!) set of philosophical questions, here, when it comes to the sacred and the profane.
As an example... have you ever heard the Johnny Cash song Hurt? It's fairly well-regarded -- not unusual to see in top fifty of all-times list -- but if you've not, it's worth listening to. It's a haunting song of depression, self-destruction, and mistaken choices that still can bring a tear to my eye; while I'm not especially attached to the genre, the singer takes the genre to the limits of its emotional range. Shelly's Ozymandias put to song, in a way, and made more impactful by how its framing interacts with the mortality of the leads and even its setting: the singer and his wife, who feature heavily in the music video, were already in poor health at the time of recording, and died not long after, while the abandoned Johnny Cash museum that they perform in would burn to the ground EDIT: would get turned into offices and a cafe within the decade, and the singer's mansion would burn down a few years after the song's release.
I don't have much interaction with the sacred, but that's pretty close, for me.
There's many other versions, as one might expect for such a popular song, and while some feel very much like they're using the song rather than treating it respectfully -- Rick and Morty used it as a season finale closer, in the same sense that Shrek used 'Hallelujah' -- perhaps the best-known is the Nine Inch Nails version. It's not bad, from a genre and technical sense... but it feels profane, compared to the Cash version. Part of that's a matter of context: a seventy-year-old Cash's needles aren't the same as a thirty- or forty-year-old Trent Reznor's. Some of that's just that the gimmicks Reznor's video uses (an atomic bomb, a decaying animal played in reverse) happened to become dated where Cash's didn't.
But the overarcing piece is just so heavily opposed to the themes of the Cash version that it's jarring. Cash's version starts slow and gradually builds across the entire piece to its final crescendo, before the inevitable fading conclusion. NIN clamors cymbals throughout points, sometimes interrupting or overriding the lyrics and the rise and fade of action. Razor used the song as an opener, and the music video (and at least some radio cuts) end in applause.
There are two lyrical differences: Cash focuses on "the pain" and wears a biblical "crown of thorns", NIN focuses on "my pain" and dresses its lead in "a crown of shit". Cash's song is a ruler mourning the rampage of time and unavoidable mistake across an empire that no longer even remembers its once-master, resigned that even could the man regain his station, that he can not help but hurt those who still care for him with his own death. Reznor's vocals are angry, a drug addict searching for the next high, driving away everyone he once loved, moaning all the time about those hurts and knowing errors, offering and threatening anything for that next rush. It's not that the Reznor version is wrong, but it's coming to a deep subject without the earnest seriousness you'd expect or hope.
Of course, Cash's cover came seven years after Reznor's original.
/needle scratch/
That's an extreme example. Reznor thought Cash made the song more of his own than NIN had, and Reznor had hesitated to allow the cover precisely because it was so close to Reznor's heart. But I have no doubt that there are people that think of NIN's message as Sacred in the way I react to Cash's -- someone's pain being driven by addiction and hubris and anger makes it no less real.
((Presumably, someone that likes The Magician's.))
I'm sure, as well, that there are versions that go the other direction : Christian or 'Christian'-themed songs that only took a serious effort at resonating with virtue when put into other framework (Cohen's 'Hallelujah' was actually popularized by John Cale... and Shrek), and I expect the majority of Sacred works are more sacred in their original tellings to their original audiences than the shoddy repackagings.
Which is a long story to say that this feeling exists, and it doesn't necessarily mean we've got to respect it... (and even argues against, to some extent)
But there's an awkward bit, there. We have decided that we're going to respect some sacred matters. Indeed, there's a pretty sizable list: socons might mock them as 'hurt feelings', but whatever you call it, there's a wide variety of discomforts where we allow massive social force and, in many jurisdictions, employment impact and direct legal impact. Yet it's hard to have this conversation without mentioning Serrano: it's not just that the profane must be allowed, but that it must be accepted and the state actively funds it and its shallow pretenses. Or to contrast varying responses to different sorts of public statue iconoclasm.
I don't know that this should fit in this category. It's very easy, as socons point out, once offense is a tool, to make being offended your core. I don't think a lot of the people raising objections about this care about flutes, or twerking, or the Library of Congress. The closest parallels I can think of -- protests being photoshopped least they offend politicians and children, at the risk of repeating myself, the outrage over the McDonald's meals at the White House -- aren't quite the same.
But this dismissal seems like a failure to engage with the problem. The Beatles did their Budokan show in 1966. Nevermind that most people here don't know anything about traditional Japanese martial arts competitions, or that the venue did get transformed into a music hall and pro-wrestling show house; none of us know this Sacred nature, if it exists, and very few people here would be old enough to have been exposed to it even had we been born in immediately-post-WWII-era Japan.
The steelman is that Britain (and Massachusetts, and New York, and Philadelphia) separates having knives from carrying knives: where the former is allowed (with a whole bunch of obscure rules that are easy to violate and have no sane rationale), the latter is prohibited. The distinction is not primarily the weapon, but treating the blade as a weapon, not just in use but in intent. Britpackers aren't automatically in trouble should they pick up a breadknife -- indeed, you'll get one in a napkin roll with your katsu chicken sandwich at the pub -- it's when they act as though this is a tool to harm other humans (or in some cases, animals), that this becomes an act which can throw you into the slammer. You're stuck accepting a bobby's or judge's interpretation for where this line actually falls, but welcome to the UK.
((Though these rules are prone to change; what style of knife you can buy in the local shop or would receive in a napkin roll isn't some law of physics.))
In the United Kingdom, this goes under the framework of "good reason", for reasons that a lot of US people following Bruen will find hilarious. But Canada is in a similar boat, here:
And for certain frameworks of that whole "state as monopoly on lethal force", this kinda makes sense. Self-defense is permitted in both countries in some circumstances, but that doesn't require the state allow people to go out prepared to survive a fight. After all, they could just abandon the public streets if they absolutely knew those threats existed, and shouldn't need weapons if those threats don't.
It's just an absolutely appalling conclusion, to my norms, not a wrong one.
More options
Context Copy link