Does this level of detail seem strange to anybody else? Shining such a strong light on OpenAI's moderation/manual review efforts seems like it would draw unwanted attention to the fact that ChatGPT conversations are anything but private, and seems somewhat at odds with their recent outrage about the subpoena for user chats in the NYT case.
Manual reviews of sensitive data are ok as long as their own employees are the reviewers, I suppose?
I thought that was pretty open? Even their more privacy-oriented Zero Data Retention agreement (which isn’t so easy to get on your business account) includes an exception “where needed to comply with law or combat misuse”
that creepy feeling of "being watched" has mostly kept me from taking advantage of any SOTA models, i only dabble in a few local ones.
The level of detail does not seem surprising. they're both charged with maintaining a facade of privacy while eliminating any and all miss-use. Certainly they heavily analyze basically everything given to them.
And generally as a society we've been ok with basically zero privacy as long as the data we send stays inside the company we sent it too. Google reads all your emails? Sure thing, read away, just don't send them to the popo. Apple knows when you're ovulating? no problem, just don't tell Amazon. etc
Same here. My assumption is that anything sent to a hosted model is public information, it will be trained on and it will be collated with your identity. And even if public models have guardrails that prevent that information from being regurgitated as slop, every CEO/owner/investor/government/etc will all have access to the uncensored models that include everything.
If you've ever ran a SaaS business, you know this and you know you can have "God Mode" access to everything, even if you swear up and down that you don't/won't.
The owners of these models aren't your friends, they see you as objects. They want to take as much value as they possibly can from you and will starve you if/when the option appears. That includes selling and sharing whatever data they have on you to the highest bidders, and some of those bidders want scapegoats to parade around as domestic terrorists.
The fact that companies are willing to send their IP and business processes to entities that can easily launder it and out compete them is mind-boggling, as well.
If you own a store and people walk in, you observe this and take a mental note. You know who visits the store. If you sell tokens, you know who buys and what people buy. It's just that now, the metadata (what I buy, when I buy, what I look like) and my intrinsics (my data) are one. I send tokens, get tokens back. If there was a way to round-robin somehow across vendors to control who gets what, I'd do it.
When contracting out manufacturing, it's common sense to spread across manufacturers, so no single manufacturer has everything. They may have half a shell. Or an peripheral module without the core. Or a core without anything around it.
Get list of your inputs mixed with generated ones and ask some model to tell you which ones are yours.
Other than that the approach in general is weak, most people likely generate lots of noise themselves. It's just about that one time you asked about X.
I was in Shanghai recently and while casually testing one of their AI chat bots I typed "What do you think of the situation in Taiwan?".
It started discussing like a Western bot would - "it's complicated, etc. etc." and around 5s it abruptly stopped and regurgitated the same line the CCP uses "... it's an unalienable part of China etc. etc.".
After printing the line, a popup opened and my camera was activated. The app wanted me to submit my information, presumably to decide what to do with me next time I enter China.
1) All the lights and modern buildings cannot hide that China is a creepy authoritarian state underneath.
2) Given the bot started printing the Western consensus first, I bet $10 it was trained by distilling ChatGPT or Gemini.
> After printing the line, a popup opened and my camera was activated. The app wanted me to submit my information, presumably to decide what to do with me next time I enter China.
Was this on your personal device? I'm just wondering how it activated your camera. I would love more details!
If this were true, why didn’t the chatbot immediately recognize that the word “Taiwan” should trigger the response? Detecting the word “Taiwan” has been possible since before most of us were born.
China has more restrictions on what you can say than the U.S. but what you are describing is not reality. Some westerner asking Deepseek about Taiwan is completely uninteresting. Just as the government do not chase people over VPN usage.
China doesn’t try to hide that they are an authoritarian state. They don’t need to. Most people in China are no less happy with their government than westerners are with their governments. Governments reflect culture. And as for foreigners, our view of China is far worse than it actually is, China doesn’t need to hide anything, people who visit China will come away with a more positive view of the country than those who do not visit.
> If this were true, why didn’t the chatbot immediately recognize that the word “Taiwan” should trigger the response?
Not recognizing they were outputting wrongthink until after it was being streamed to the user is a known behavior with some Chinese chatbot apps. A quick search found an example of DeepSeek doing it: https://www.reddit.com/r/OpenAI/comments/1ic3kl6/deepseek_ce...
I don't think his story is genuine, but it showing the "wrong" answer before correcting itself is known behavior.
I've seen it from the non-Chinese ChatGPT before. Something was deemed to be violating the sensitivity filters or something, and it refused to answer. But only after I saw part of the real answer streamed to the output, and then redacted and replaced.
My wife grew up in Shanghai, and you'll have to go quite some distance to find someone more critical of the PRC and CCP than she is. And it's with good reason.
She grew up during the cultural revolution, and was largely raised by her grandmother because literally every other person in her extended family was in prison or work camp, not because of anything they had actually done wrong, but for political reasons because the whole family was blacklisted.
And that's not just the old days. Her father died as a direct result of Chinese Covid policy. During the pandemic her cousins still in the country would ask her (on Skype) "is X true?", and largely their perception of what was going on was false. She would exfiltrate encrypted news reports to them - until those started getting blocked. Her dad's estate still has affairs that need to be resolved, but we've decided not to return to China until Xi is gone, as it's just not safe. It doesn't get much airplay, but there are currently a couple of hundred Americans who are being illegally detained in China right now. It's not worth the risk.
My first trip to China was about 30 years ago, shortly after we got married. And back then, I would have said that you were right. Honestly, it felt like for the average person in their day-to-day-lives, the Chinese were less under the governmental thumb than we are. People from the countryside would bring their produce into the city to sell, or cook dumplings and buns to sell on the side of the street - stuff that in America we'd have to get permits for. It seemed that the oligarchy had an understanding with the people: let us control the big picture, and we'll look the other way for the little things. But Chinese politics is a pendulum swinging very widely. From Tienanmen Square and Tank Man, it had swung quite a bit the other way. But today, it's come back 180-degrees. Xi is really trying for a Cultural Revolution 2.0.
These impressions largely match what I hear from other Chinese immigrants - except for Party members, who tend not to want to talk about it at all. I'm afraid that you've been listening to too much propaganda.
i don't doubt your experience, but just know it might be skewed and not representative of everyone's opinions
the sense i get from my chinese friends are that the CCP is an annoying parent but they understand the challenges both domestic and international and largely agree with the compromises
How do they feel about and respond when asked about the Taiwan question?
Do they either clam up or act like it's a mortal insult to suggest that an independent democratic nation should not live in fear of impending violent conquest?
Because that's the kind of reaction that makes the reports of "happy life, all's good" a little harder to digest.
Not saying that's a unanimous opinion / response, of course. But it certainly seems to be the default.
About 50% of Chinese people I meet very much agree with the government that it's part of China and always has been. The other 50% know that it's clearly independent and are tired of the whole act by the Chinese government. But the people I've talked to about this are people with the means to travel, and many of them have been to Taiwan. So it may not be representative of the typical person on the street. I've been to China several times and I don't want to ask it there, but that's less out of fear of the government but more than I don't want to bother locals with politics and present myself as an enlightened foreigner, since nobody likes that shit. Just like nobody would like a Chinese guy going to Alabama and telling the people they need to embrace socialism if they ever want to escape poverty.
The majority of US support for Taiwan and it's current situation is owed entirely to supporting a military junta from the mainland that massacred the local Taiwainese who objected to it and suppressed civil society.
Are you saying you would've been neutral on an invasion of Taiwan before 1985 or so, since it wasn't a democracy?
Cultural Revolution is all about totally politicalised society, extremely polarised, regular people fight against each other based on ideologies. Isn't that the current west?
> because literally every other person in her extended family was in prison or work camp
translate for you - her family was heavily involved in politics, it is just unlucky that her family was not on the winning side, so she hates whatever happened.
posting from Shanghai, going back to the 3rd world west in a few days.
translate for you - her family was heavily involved in politics, it is just unlucky that her family was not on the winning side, so she hates whatever happened.
This is false. When you have no idea at all what you're talking about, you should just be quiet.
The problems were that (in order of increasing specificity)
(a) We're talking about Marxism here, and Marxism is all about class warfare. Before the Communists her family had been part of the "landlord" class, and thus were enemies of the people by definition.
(b) One uncle was tricked by the anti-rightist movement. If you're not aware of this, it was earlier in Mao's reign. Mao said, essentially, "we know we haven't gotten everything perfect, so tell us what we could do better". Wife's uncle was stupid enough to believe him, the result of which was a 20-year prison sentence, and also his wife being forced to divorce him, and further tainting the family. (Something on the order of 500K to 2M people were persecuted like this.)
(c) Any outside influences were suspect at best, and often de facto proof of espionage. She had an uncle who was a US citizen. And her father had traveled extensively internationally, as a sea captain (never mind the fact it was the PRC government, as the sole employer in China, who put him onto those ships).
(d) Wife's family side had been in theater. One aunt had been in a theater troupe with Jiang Qing (Mao's wife), and knew at least some of her, ummm, lower class history. Putting her, and the rest of the family, in prison kept them shut up and warned them not to talk any further. (This may sound far-fetched, but consider what the Gang of Four was up to during the Cultural Revolution, and that she was a member.)
posting from Shanghai, going back to the 3rd world west in a few days.
You might do well to read, e.g., Shanghai Tears by Pu Gui Yuan to better understand what was happening back in those days. Then again, I don't imagine you can just go buy a copy of it over there.
No. But there are actual circumstances here that differ between China's actions and the rest of the planet. Specifically...
While the rest of the world was doing stuff like ensuring that as many of its citizens as possible were vaccinated, and letting the population gradually work up to herd immunity so that controls could be gradually loosened, China kept the population at a hard lockdown right to late 2022, and then opened up completely. It was as if they just opened the floodgates.
There were actually people arguing that China was doing this intentionally, with the plan being to thin out the top-heavy aging demographic in the country. I'm not necessarily advocating for this theory, but illustrating that the very fact that there's a colorable argument for it demonstrates how irresponsible Chinese leadership were.
The result was that in my father-in-law's retirement home, literally EVERY caretaker came down with the virus together, which obviously led to most of the residents getting sick. And given the way covid worked, that meant a whole lot of deaths.
Adding insult to injury, his death certificate attributes the cause of death to heart disease. As a matter of policy, all deaths were attributed to any other condition the patient might have had, however trivial, unless covid could be proven. And proving it would involve in declining to properly dispose of the body, paying for the autopsy and so forth. But there's no doubt (having talked to him every day on Skype) that covid is what killed him.
China's policy reduced the death rate by a factor of about 75% relative to the US.
The zero-Covid policy kept Covid out of the country until an effective vaccine was developed and deployed to about 90% of the population. The main problem was that there's a widespread belief in East Asia (including in Taiwan, Singapore, etc.) that vaccination is dangerous for old people, so the vaccination rate was lowest among the most vulnerable group. A lot of old people simply refused to get vaccinated, despite large vaccination drives and public messaging asking them to do so.
Then, as you said, the zero-Covid policy was eliminated overnight, and practically everyone in the country got Covid within 1-2 months. However, because most people were vaccinated, the death rate was far lower than in the West.
All in all, the zero-Covid policy saved several million lives in China. This is based on retrospective studies by outside researchers, not on official statistics.
DeepSeek would print all it's mental gymnastics to censor itself in the reasoning phase directly to the user, before shutting down the conversation. Apparantly such an odd move is a thing in China.
Right, I think deepseek continues to be massively misunderstood. It appears to be a replication of existing technologies done more efficiently rather than a breakthrough in terms of bootstrapping from the ground up with new capabilities. And at this point people will start saying "well does that matter?" and the answer is yes.
>And as for foreigners, our view of China is far worse than it actually is, China doesn’t need to hide anything, people who visit China will come away with a more positive view of the country than those who do not visit.
To the extent that's true, it's because they won't let you see the uyghur reeducation camps.
We can get videos from remote hellholes of Africa like Dafur and Mali but apparently,that's too much to ask in Xinjiang.We can't even get satellite images to show us evidence of this so called wigur genocide
On the other hand you can travel to Xinjiang, visit mosques, Uighur museums, experience Uighur culture, observe Uighurs just minding their own business in their daily life.
“Subjected to arbitrary arrests and forced labor, sterilizations to torture, more than one million Uyghurs, Kazakhs, Kyrgyz, and other minorities are estimated to have been locked up in so-called “re-education” camps and prisons in the region over the last decade, according to the UN Office of the High Commissioner for Human Rights.”
I ran an anonymized Facebook account for years with thousands of followers that mainly sticks to news and politics.
Once I started criticizing Libs of TikTok, the propaganda arm for this administration, and getting traction with users, my account was locked and now I have to scan my face and ID if I want to use it again.
>Given the bot started printing the Western consensus first, I bet $10 it was trained by distilling ChatGPT or Gemini.
To your point I've seen something similar with Deepseek, generic answers start printing and then, in plain sight, removed and replaced with a non committal message along the lines of "I don't have access to that information."
What if I say I AM a Chinese netizen, right here in China Mainland talking to you?
What if I say no application on my phone ever turn my camera on without my prior approval? What if I confidently say the data privacy situation in China is not in any way worse than USA?
You say I'm censored by gov. Yeah, and so do YOU. We are quite the same, so don't laugh at each other.
If you ever allowed the app access to your camera then next time it will be able to access your camera.
No one is suggesting the app need have circumvented standard android permissions. I'm saying it is not surprising the app would try to open the camera on violation of rules.
You might be surprised to know that I do not disagree with what you said. When it comes to data privacy your information is safe from other private companies but not the government.
> You say I'm censored by gov.
When it comes to political inquiry? Without question this is more sensitive in China. My point is that in China apps are coerced to reveal information about their users to a degree where actively trying to take a picture would not be surprising, or much of an escalation. Not that it would be needed anyways since Internet usage is tied to ID.
I'm pretty sure they can just prompt any convo in the background and ask "is this conversation sensitive ?" and the model can answer without this being added to the context of the convo.
One hopes the CIA/Secret service would be willing to provide the human to do the reviewing but sadly I've worked for European telco's and I know better.
The amount of information about everything that people are giving OpenAI is astronomical, information that was previously kept closely guarded is now just freely flowing through foreign servers.
Truly a paradise for american intelligence. Would have expected that the chinese officials be briefed on not using us tech companies, but opsec is hard to teach, and even harder to always follow.
I think one of the reasons why AI companies are valued this high is one can actually inspect what user inputs & outputs are.
It's basically an OSINT siphon.
In this Chinese case, the tokens are leaked at least twice. ChatGPT offers no direct access to the Chinese, they have to use some kind of Openrouter-like service, but the data where also in clear-text during transmission.
> While the US certainly isn't pure, its scale of surveillance intrusion is light compared to China
I assume that for someone to believe this, they either have to believe the U.S. has poorer surveillance capability than China, or, more likely, they consider U.S. surveillance unintrusive and Chinese surveillance intrusive.
> ... or, more likely, they consider U.S. surveillance unintrusive and Chinese surveillance intrusive.
Of course. What's the point of surveillance if you're not going to use it to enforce dogma? I think you can reasonably evaluate a country's surveillance by looking at the pettiness of the arrests & censorship they make.
See this chinese tech reviewer[1] being bullied by the government for putting a spotlight on chinese phone makers cheating about benchmarks. I'm not sure the US is at this point yet...
> See this chinese tech reviewer[1] being bullied by the government for putting a spotlight on chinese phone makers cheating about benchmarks. I'm not sure the US is at this point yet...
Plenty of signs point towards this being a case where it wasn't the government, so it's not a good example. Instead it looks to have been the tech companies filing takedowns and threatening lawsuits. Especially when it comes to China, this is a big difference.
You should've pointed towards more clear-cut examples like Naomi Wu and Peng Shuai. But those cases are less unthinkable in the US, and it should be uncontroversial to say that it's what's been worked towards.
If I were doing this sort of thing, I would make certain to ban accounts that were too obvious while leaving ones that are subtle enough, so that the other side has less reason to suspect I am tracking their inputs and feeding them disinformation.
This tells us that we should never share sensitive information with GPT, even if you’ve set it not to use your data for training. Nothing can stop OpenAI from misusing your data.
> “This is what Chinese modern transnational repression looks like,” Ben Nimmo, principal investigator at OpenAI, told reporters ahead of the report’s release. “It’s not just digital. It’s not just about trolling. It’s industrialized. [...]
There's something poetic about OpenAI being asked to comment on mis-use of their slop generator, and their answer is composed entirely of AI slop.
lol everyone claims deepseek and all chinese companies are collecting private information and ban them in western companies. but it is okay being spied by openai :)))
I like DeepSeek because of their pricing, although I'm still evaluating. I wonder if I'll need a VPN in the future to access it though (from EU). Cheap is good, cheap prevails.
The official DeepSeek API is routed through AWS load balancing btw.
i kinda get the impression this was from 2023 and also it is not clear what this dissident did, hard to evaluate whether i should care without knowing that
> intimidating Chinese dissidents abroad, including by impersonating US immigration officials
I hope those victims of immigration impersonation don't have family within China's borders. AI-enabled impersonation and intimidation are far from the worst of China's crimes [1] against its overseas critics.
China likes to make you an offer you can't refuse [2] [3]: You're saying stuff the Chinese government doesn't like, but you live outside its borders and the secret police can't get at you? You need to come to China and be jailed (or worse). If you don't, your family will be the ones who are jailed (or worse). Or you can unalive yourself, and save the glorious Chinese Communist Party the expense of a bullet.
[1] China would say "the government punishes a criminal's family" is not a crime, it's a perfectly legal implementation of government policy under Chinese law. I respond that the death camps were perfectly legal implementation of government policy under Nazi law, but were still crimes against humanity -- China's actions fall in this category of crimes.
As I understand it: Western societies have a very individualistic view of responsibility. If you didn't commit a crime, you're innocent. Punishing the innocent family members of a criminal is morally abominable.
In the Chinese Communist Party's view, criminal responsibility is collectivist. By their definition, the family members of a criminal share responsibility for the crime regardless of their participation in the criminal acts. "Innocent family members of a criminal" is a logically inconsistent concept in their world view. The family of a criminal is guilty by definition -- being related to a criminal is itself a crime.
> “It’s not just digital. It’s not just about trolling. It’s industrialized. It’s about trying to hit critics of the CCP [Chinese Communist Party] with everything, everywhere, all at once.”
Does this level of detail seem strange to anybody else? Shining such a strong light on OpenAI's moderation/manual review efforts seems like it would draw unwanted attention to the fact that ChatGPT conversations are anything but private, and seems somewhat at odds with their recent outrage about the subpoena for user chats in the NYT case.
Manual reviews of sensitive data are ok as long as their own employees are the reviewers, I suppose?