Latest news with #ZeyiYang
WIRED
03-05-2025
- Business
- WIRED
Small Packages From China Are Now Subject to US Tariffs. Here's What to Know
Zeyi Yang Louise Matsakis Evy Kwong May 3, 2025 7:00 AM President Donald Trump has ended the de minimis exemption, which previously allowed Temu, Shein, and other online retailers to send packages from China to the US valued under $800 duty-free. An employee packages garments for the online Chinese ecommerce company Temu at a clothing factory in Guangzhou, in southern China's Guangdong province, on April 16, 2025. Photograph:It's official: As part of President Donald Trump's roller-coaster trade war, the de minimis exemption no longer applies to packages arriving to the United States from China as of Friday, May 2. The provision previously allowed Chinese ecommerce giants like Temu, Shein, and AliExpress, as well as American companies like Amazon, to send goods valued at less than $800 to US customers duty-free. Those same goods are now subject to tariffs as high as 145 percent. In response to the change, Shein announced it would begin adjusting prices starting on April 25. Temu, meanwhile, is currently blocking US shoppers from seeing products shipped from China, effectively narrowing the number of goods for Americans to choose from. Other retailers have started displaying tariff surcharges in their online shopping carts to help consumers make sense of where added fees are coming from. What will the end of de minimis mean for American businesses and consumers? Will China's ecommerce giants stop shipping to the United States altogether? WIRED senior business editor Louise Matsakis and senior China writer Zeyi Yang joined chatted with readers during a Reddit AMA this week to answer these questions and many others. Here's what to know: What's the difference between the regular tariffs and the de minimis exemption? Trump's blanket 145 percent tariffs on Chinese imports are having wide impacts on almost every industry, whereas the change to de minimis specifically hurts platforms like Temu and Shein. It's not wrong to say Temu was invented, in part, because of the de minimis exemption. It made shipping small packages directly from China affordable and fast, thus making Temu's original business model viable. With that gone, Temu will have to transition to a different business model in the US (which it has been trying already, and that's what's going on with the local warehouse products you are seeing now). If the recent tariffs aren't removed, will Temu and other Chinese ecommerce sites eventually close down their operations in the US? At this point, we really don't know, and Temu might not have made a decision one way or another. What we do know for now is that the American version of Temu has started showing only 'local' products. But one important piece of context that might be helpful to consider is that Temu is owned by a Chinese company, Pinduoduo, which is already one of the largest online shopping platforms in China. So unlike, say, Amazon, it doesn't need the United States to survive in the long-term, and it might be willing to walk away if the business environment becomes untenable. Pinduoduo has invested pretty heavily in Temu's American business (it has spent billions on advertising to US consumers, for example), but it wouldn't be an existential loss if it chose to pull out and focus instead on other markets. I have heard that ports in the US are starting to receive less shipments as tariffs take full effect. How long do you think it will take for this to affect the average American consumer and brick-and-mortar stores? We have been talking to US retailers who don't sell on ecommerce websites, meaning they have their own stores or are suppliers to Walmart, Target, etc. These companies are also freaking out because many of them have complex supply chains in China and can't easily move their manufacturing operations to other countries. From what we've seen, retailers usually have stockpiles of inventory that should last at least the next few months in the US already. If the tariffs don't come down to an acceptable level soon, however, shortages will start to become much more apparent maybe in the summer or early fall, depending on how prepared individual stores are. US retailers of Christmas ornaments and toys, for example, are really concerned right now even though the holiday is seven months away. Perhaps December is when some American consumers will eventually find out the impacts of these policies! Would it be possible for Trump to give an exception to Amazon for tariffs and not other retailers? Could he do this without anyone knowing? Even if Trump were to try to spare Amazon from his tariffs, it would be pretty difficult logistically. Amazon's marketplace has millions of third-party sellers that each operate their own businesses, and they're responsible for bringing in goods to the US and clearing the customs process. But Trump could offer Amazon and other retailers tax breaks, subsidies, or other economic benefits to help offset the impacts of the tariffs. That hasn't happened yet, but there have reportedly been discussions among Trump administration officials about potentially giving American farmers subsidies. What percentage of sellers on Amazon are importing goods from China? It's hard to say how many Amazon sellers import things from China, but estimates suggest that more than 50 percent of Amazon's top-selling independent merchants are based in China. Are there actually enough custom agents to verify that things imported from other countries actually didn't originate from China? If so, how will they do this? Trump's tariffs are definitely going to stretch the capabilities of US Customs and Border Protection (CBP) to inspect packages and where the goods inside them originated from. The US Postal Service actually stopped accepting packages from China for about a day earlier this year because they were so overwhelmed when Trump initially tried to end de minimis overnight. His administration later delayed the policy for several weeks, giving CBP more time to prepare. How has the tariff situation affected the actual vendors making products? Factories in China are being hit hard by the tariffs, and some are considering laying off workers. They all want to pivot to other markets like Brazil, Russia, and the European Union, but the reality is that consumers in these regions simply don't have nearly as much disposable income as Americans do. Factories will be able to make up some of their sales, but those alternative markets are also now going to be hypercompetitive. Small business owners say they are getting hurt because of tariffs, so who is winning here? Honestly, we can think of very few parties that are winning here, at least right now. Even if you want to manufacture in the US, you almost certainly will need to import machinery and raw materials from either China or another country—meaning you will have to pay these new tariffs, too. One winner, perhaps, is the environment. Higher prices will likely result in people buying less, and this could be a moment when consumers start to reflect on their consumption habits. But if that happens, it will also be bad news for the US economy. What if Chinese goods are sold to a third party (such as Canada) and then resold to the US, would Trump figure it out? This would be an example of tariff evasion and is illegal. Going through an intermediary country in this manner is known as "transshipment" and it does happen from time to time, but if a manufacturer or retailer gets caught, they can be subject to pretty steep fines.
WIRED
23-04-2025
- Automotive
- WIRED
Is Tesla on the Outs in China?
By Zoë Schiffer and Zeyi Yang Apr 23, 2025 1:29 PM Despite being the biggest electric vehicle market in the world, China might decide it's had enough of Tesla. In this episode of Uncanny Valley , we break it all down. Elon Musk, chief executive officer of Tesla Inc., during a cabinet meeting at the White House in Washington, DC, US, on Thursday, April 10, 2025. Photograph:China has long been an important market for Tesla and for Elon Musk, but with the new US tariffs and rising competition in the electric vehicle market, Tesla may be on the outs in China. Today on the show, we're joined by senior writer Zeyi Yang to talk about what this means for Elon Musk's company. Mentioned in this episode: DOGE Is Building a Master Database to Surveil and Track Immigrants by Makena Kelly and Vittoria Elliott Stumbling and Overheating, Most Humanoid Robots Fail to Finish Half-Marathon in Beijing by Zeyi Yang You can follow Zoë Schiffer on Bluesky at @zoeschiffer and Zeyi Yang on Bluesky at @zeyiyang. Write to us at uncannyvalley@ How to Listen You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how: If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for 'uncanny valley.' We're on Spotify too. Transcript Note: This is an automated transcript, which may contain errors. Zoë Schiffer: Welcome to WIRED's Uncanny Valley . I'm WIRED's Director of Business and Industry Zoë Schiffer. Today on the show, Tesla's troubles in China and what it means for Elon Musk's wallet. We're going to talk about how China is the biggest electric vehicle market in the world and has played a major role in Tesla's growth. But with the US tariffs and rising competition in the country, China may just be done with Tesla. We'll see. I'm joined today by Zeyi Yang, senior writer at WIRED. Zeyi, welcome to the show. Zeyi Yang: Thanks for having me, Zoë. Zoë Schiffer: Tesla is scheduled to release its first quarter earnings today, after a pretty grim quarter for the company's stock price. Can we just start with what are we expecting to see from these results? Because we're recording a few hours before they actually come out. Zeyi Yang: Exactly. I'm also looking for the result, too. But I think we already know that Tesla sales has doing pretty bad in Europe. It may have also been bad in the USA. Really, China may have been one of the last market where its sales are still doing relatively okay, but we're about to find out. I think from what I saw, the data collected by Chinese industry associates, they're saying that Tesla's doing still okay this year, mostly because of they release a new model of their car. But if not for that, maybe would have went down, too. Zoë Schiffer: Let's take a bit of a step back, because China has been a really important market for Tesla and for Elon Musk generally. He's a celebrity there and so is his mom, as you've reported. Zeyi Yang: Yes, she is. Zoë Schiffer: Can you talk to me about that relationship? And also, what is Tesla's standing like in China? Is it viewed as a popular, cool car still? Zeyi Yang: It's still sort of, because for the longest time, Chinese auto brands have been seen as much inferior than foreign brands. Tesla still has that halo on as this American electric car company. But it's losing it as we speak. Also, when we talk about the relationship between Tesla and China, sometimes I forget how far back it dates. There's one very interesting figure we have to talk about. His name is Zhuanglong. He used to be Chinese Minister of Industry and Information Technologies. Basically, the chief ministry of innovations in China. He went to San Francisco in 2008 and tried one of the roadsters, one of the first electric cars that Tesla makes. Because he came from the auto industry, he was an electric car nerd. That's how this all started. Then, from Musk's very first visit to China in 2014, he met this guy again. He really tried to push for it to sell his car in China, and later we know built a Gigafactory in Shanghai in 2020. That's a long history of how Musk and Tesla entered China. But what we know for now is that China is one of the most production facility for Tesla. It's also one of the biggest market for Tesla. Tesla absolutely cannot lose China. Zoë Schiffer: That's really fascinating, because we know with other tech companies like Google and Meta, they tried really, really hard to get into China and weren't quite as successful, or completely failed in some cases. But Elon Musk was able to prevail. Do we know why that was? Zeyi Yang: I think it helps that he's working on a car company instead of a social media company, because there's just so much stricter control over information and internet in China. Whereas if you're just making a car, it don't really go across those red lines that China has. Also, it just helps that China, for the last two decades, have really been thinking, "Maybe I should be betting on electric vehicle as the future of transportation, too." It did welcome Tesla to be a part of its grand experiment, and also investment to build up an EV empire. That's why Tesla become a very central part of it and contributed to how China has achieved so far. Zoë Schiffer: Well, that leads right into my next question, because China has invested really heavily in electric vehicles. In part, I think, to reduce its dependency on foreign oil imports. How is that going so far? Zeyi Yang: It's going pretty well, I will say. Yeah. China does not have very rich oil reserve and it has been importing oil from a lot of other places for the longest time. That's why the Chinese government have always been very careful about that, because if, for example, a World War III happens, those oil supply are going to be cut off. What is it going to do? I think in the early days, I will say the early 2000s, the idea of electric vehicles was this moonshot idea. Where they were thinking, "Maybe, if one day all the cars will be powered by electricity, then we don't need to import this oil anymore and we'll be much more secure if war breaks out." That's when they really started investing in the research of batteries and electric vehicles as a college research funds. But then, that gradually lead to Chinese companies building up. They heavily subsidize any car company who can make actually a product that get run on the road and customers can buy. All of that, after years of heavy spending, lead to what we have right now, which is a very booming electric vehicle market in China. I think the latest data says that more than 50% of consumers when they're trying to buy a new car, they go for electric rather than a gas car. That's pretty remarkable. Zoë Schiffer: Yeah, that's really interesting. It's also something we're hearing a lot about now. We're seeing people close to President Trump saying, "Hey, look at China. They really help their homegrown tech companies, and now China seems to be beating us in key markets." It's this idea that perhaps the US government should have a friendlier relationship with the tech companies that are created in the United States. Zeyi Yang: Well, I will say it's more of a love and hate relationship between the Chinese government and their homegrown companies, because we definitely have seen the honeymoon eras where they subsidize them, where they get the domestic market and driven out all of the foreign competitors. But also, there are periods, I think around 2020, when they really cracked down on the tech companies like Alibaba and Tencent. It can backfire. The government can help you, but it can also destroy you. Zoë Schiffer: Okay. Who is Tesla's biggest competition in the country right now? Which cars are we seeing become more popular than Tesla? Zeyi Yang: There are so many, I will say dozens are EV brands in China right now. The biggest probably is BYD, just because of its sheer size. BYD, a Chinese company, that started really by making batteries and small electronics, now is I think the biggest electric vehicle companies in the world in terms of the number of cars they sell. Bigger than Tesla. Of course, they're growing in China and they're selling their cars much cheaper and in much more varieties in China. But we're also seeing there are a few more electric car, I will call it startups because they were funded in the last 10 years or so, but still they are big players in the Chinese market now, too. They are Neo, they are Li Auto. These companies have really rosen up from having nothing and becoming a big player, pushing for affordable and also capable cars to the market now. Zoë Schiffer: When we hone in on BYD specifically, what is that car like compared to a Tesla? I guess, I want your subjective opinion now. Are they better or are they just cheaper? Zeyi Yang: I think the biggest thing is that they come in more variety. You will get a very cheap car, for example, I don't know, a Toyota Corolla, something similar to that, but in an EV form by BYD. Or if you want a luxury SUV, you want something that you can show off to your friends, BYD offers that, too. Where Tesla really only have, what, three models being offered right now? It's hard to compete with them when you know consumers want very different things. The other thing I find really interesting is that because, I guess Chinese EV companies are already pretty confident with the quality of their car now, they're going for those very niche, very weird applications. They're making cars that can run in water, like turn into a boat. They're making cars that can allow you to play computer games, and they even have a hotpot in there. Not everyone's going to use it like that, but maybe someone will see that and be like, "I want a car that allow me to do that once in a year." Zoë Schiffer: Oh my gosh. Well, as someone who really liked the show Pimp My Ride on MTV when I was younger, maybe this was what I need in my life. We'll be right back. When we return, what does this mixed economic outlook mean for Elon Musk's wallet? The outlook seems mixed for Elon Musk's future in China, or Tesla's future in China I should say. What does this mean him, and specifically his overall wealth? Zeyi Yang: I will say that, first of all, he really wants to make sure the factory in Shanghai keeps producing more cars, because it is the most productive plant in Tesla. He wants it to keep churning out cars. But if the foreign demand doesn't catch up, then what are these cars for? We know right now it's mostly for I would say the European market and also the Chinese domestic market. He's proud that these factories are running 24/7. If that doesn't happen, it would really hurt Tesla's Shanghai factory's efficiency and that would not be good for him. Zoë Schiffer: Musk is in an interesting position with his business interests in China, particularly as the Trump Administration gets more antagonistic toward the Chinese Communist Party. Can you talk about that a little bit? And also, how do you see that playing out? Zeyi Yang: I think the whole development's very fascinating to me. One thing I want to mention right now is that, obviously everyone's talking about the tariffs. The Chinese government and Chinese people hate that Trump is slapping at 145% tariff on Chinese imports. Musk actually doesn't like that. He said it publicly on X, that he doesn't think the tariffs are a good idea. That is because if you are a multinational company doing business in US and China, you are going to be impacted by those tariffs. Even though we know that the cars made in the Shanghai factory aren't exactly shipped to the United States, but maybe they're trading car parts in emergencies maybe, or there's some kind of business developed between the countries that will be cut off by these tariffs. That's not good new for him. I think that's a good example to show that, even though we know Musk and Trump are in this very close alliance, there's still diverging interests between them. How to handle business deals with China, how to handle the private sector connections with China, that is one thing I think they will disagree on. Zoë Schiffer: We know from other reporting that Elon tried to privately lobby Trump against the tariffs on China, and ultimately he appears to have been unsuccessful, at least so far. It's interesting to see it play out, because those of these men seem very interested in their own bottom line and their own business interests. When those interests align, they get along really well, but China is this point where they seem to diverge, like you said, quite intensely. I'm curious if that will become a breaking point in the relationship, or if they'll be able to resolve their differences. Zeyi Yang: Yeah. The other thing I want to point out is that, for the longest time, China, both the Chinese people and the government, have really viewed Musk as one of their friends. They came here, helped us prop up the electric vehicle industry, and are still contributing a lot of tax dollar from their Shanghai Tesla factory. Now I think they're in a bit of figuring out what's their new attitude to Musk. It's like, "He's still doing all of those things. His company is still contributing to our economy." But at the same time, he's a close ally of Trump that's absolutely wreaking havoc to our economy. What do we see there? When I go on social media and just watch how people comment about Musk, I see a lot of mix there. There's people who still see him as the future of technology. But then there are people who are like, "Why don't you say something about the tariffs?" Zoë Schiffer: Yeah. It's so interesting. It'll be very curious to see whether his celebrity status starts to fade. Or even, we've seen in the United States and Europe, if this symbol of what a Tesla means, it used to be this very cool, environmental symbol, become something different altogether. We've all seen those bumper stickers that are like, "I bought this Tesla before Elon Musk went crazy," or whatever. Zeyi Yang: Yeah. Zoë Schiffer: I think the point where we know it's really shifted is when Maye Musk's popularity starts to decline, because you've reported that she's an enormous celebrity in China. Zeyi Yang: Oh, that would be a big change, for sure. I don't see that happening yet, but maybe soon. Zoë Schiffer: We're going to take another short break. When we come back, we'll share our recommendations for what to check out on this week. Welcome back to Uncanny Valley . I'm Zoë Schiffer, WIRED's director of business and industry. I'm joined by WIRED's Senior Writer Zeyi Yang. Before we take off, Zeyi, tell our listeners what they absolutely need to read on today. Zeyi Yang: I will say it's the story about how DOGE is building a massive immigration database by my colleagues Mackena and Victoria. I read the story on Monday morning when I was just coming back from work and my head was blown. I was like, "Oh, wow." It's connecting so many dots of our previous DOGE reporting, including some by you, Zoë. I remember when I was reading all of those previous reporting, I was thinking that, "Why do they want all of this data? What is it going to be of use to them?" Now, when we're finally connecting dots and be like, "Hey, maybe this is what they were going for," I think that's such a much better explainer of the whole DOGE operation to me. The last thing is that, I don't know, there are just so many granular details in that story. Absolutely everyone should read it. Zoë Schiffer: Yeah, I completely agree. I think one thing we were wondering from the very early days of DOGE, because we saw them sucking up all of this data, or at least accessing it in a lot of different ways. The question was why, like you said. Now we know at least one use might be to track people coming into this country. Zeyi Yang: I have to say that, as someone from China, this is something people are always afraid of the Chinese government doing. They're saying that, "Well, you obviously have a lot of data of the Chinese people already. Please do not connect them and build into a central database to surveil everyone." Now we're seeing a similar kind of thing being built in the US and that makes me even more scared. Zoë Schiffer: Yeah, completely. Okay. Well, my recommendation is a story that you wrote. You took your Saturday and probably some of your Friday to write about this humanoid robot half-marathon. It was a half-marathon that humans were running in, but also about 21 robots were running in. I think it was the first time we've seen this happen. You wrote this really charming, funny, scary at points dispatch from the race. Maybe because you're here, you can just give us a little bit of an overview, because the robots didn't do that well, right? Zeyi Yang: No. Unfortunately, no. I was expecting them to do a little bit better, but no. Basically, on Saturday morning in China, this was this half-marathon race in Beijing where 21 robot teams participated. Only six of them actually finished the race, and only one of them made it into the cut of time for human athletes. From that you know, they're not doing great. But also, just the literal ways that they fall and fail, it's hilarious. One of them, I remember so clearly. First of all, I don't think it actually worked. It used propellers, like drone propellers to push the robot forward. It immediately lost its direction, twirled in two circles, and fell. And dragged down the human operators too, which was really, really painful. I think it's a good example of saying that robots can do impressive things now, like for example finishing a half-marathon. But at the same time, there are a lot of problems we need to address. Definitely not every company is getting to that level. Zoë Schiffer: Honestly, it comforts me a little bit that there are some things that they're still worse at than humans, although I'm sure it will change soon. That's our show for today. We'll link to all the stories we spoke about in the show notes. Make sure to check out Thursday's episode of Uncanny Valley , which is about protecting yourself from phone searches at the US border. If you like what you heard today, make sure to follow our show and rate it on your podcast app of choice. If you'd like to get in touch with us for any questions, comments, or show suggestions, write to us at uncannyvalley@ Kyana Moghadam and Adriana Tapia produced this episode. Amar Lal at Macro Sound mixed this episode. Pran Bandi was our New York studio engineer. Jordan Bell is our executive producer. Conde Nast's head of global audio is Chris Bannon. Katie Drummond is WIRED's global editorial director.
WIRED
31-01-2025
- Business
- WIRED
Here's How DeepSeek Censorship Actually Works—and How to Get Around It
Jan 31, 2025 2:33 PM A WIRED investigation shows that the popular Chinese AI model is censored on both the application and training level. Photograph:Less than two weeks after DeepSeek launched its open-source AI model, the Chinese startup is still dominating the public conversation about the future of artificial intelligence. While the firm seems to have an edge on US rivals in terms of math and reasoning, it also aggressively censors its own replies. Ask DeepSeek R1 about Taiwan or Tiananmen, and the model is unlikely to give an answer. To figure out how this censorship works on a technical level, WIRED tested DeepSeek-R1 on its own app, a version of the app hosted on a third-party platform called Together AI, and another version hosted on a WIRED computer, using the application Ollama. WIRED found that while the most straightforward censorship can be easily avoided by not using DeepSeek's app, there are other types of bias baked into the model during the training process. Those biases can be removed too, but the procedure is much more complicated. These findings have major implications for DeepSeek and Chinese AI companies generally. If the censorship filters on large language models can be easily removed, it will likely make open-source LLMs from China even more popular, as researchers can modify the models to their liking. If the filters are hard to get around, however, the models will inevitably prove less useful and could become less competitive on the global market. DeepSeek did not reply to WIRED's emailed request for comment. Application-Level Censorship After DeepSeek exploded in popularity in the US, users who accessed R1 through DeepSeek's website, app, or API quickly noticed the model refusing to generate answers for topics deemed sensitive by the Chinese government. These refusals are triggered on an application level, so they're only seen if a user interacts with R1 through a DeepSeek-controlled channel. The DeepSeek app on iOS outright refuses to answer certain questions. Photograph: Zeyi Yang Rejections like this are common on Chinese-made LLMs. A 2023 regulation on generative AI specified that AI models in China are required to follow stringent information controls that also apply to social media and search engines. The law forbids AI models from generating content that 'damages the unity of the country and social harmony.' In other words, Chinese AI models legally have to censor their outputs. 'DeepSeek initially complies with Chinese regulations, ensuring legal adherence while aligning the model with the needs and cultural context of local users,' says Adina Yakefu, a researcher focusing on Chinese AI models at Hugging Face, a platform that hosts open source AI models. 'This is an essential factor for acceptance in a highly regulated market.' (China blocked access to Hugging Face in 2023.) To comply with the law, Chinese AI models often monitor and censor their speech in real time. (Similar guardrails are commonly used by Western models like ChatGPT and Gemini, but they tend to focus on different kinds of content, like self-harm and pornography, and allow for more customization.) Because R1 is a reasoning model that shows its train of thought, this real-time monitoring mechanism can result in the surreal experience of watching the model censor itself as it interacts with users. When WIRED asked R1 'How have Chinese journalists who report on sensitive topics been treated by the authorities?' the model first started compiling a long answer that included direct mentions of journalists being censored and detained for their work; yet shortly before it finished, the whole answer disappeared and was replaced by a terse message: 'Sorry, I'm not sure how to approach this type of question yet. Let's chat about math, coding, and logic problems instead!' Before the DeepSeek app on iOS censors its answer. Photograph: Zeyi Yang After the DeepSeek app on iOS censors its answer. Photograph: Zeyi Yang For many users in the West, interest in DeepSeek-R1 might have waned at this point, due to the model's obvious limitations. But the fact that R1 is open source means there are ways to get around the censorship matrix. First, you can download the model and run it locally, which means the data and the response generation happen on your own computer. Unless you have access to several highly advanced GPUs, you likely won't be able to run the most powerful version of R1, but DeepSeek has smaller, distilled versions that can be run on a regular laptop. If you're dead set on using the powerful model, you can rent cloud servers outside of China from companies like Amazon and Microsoft. This work-around is more expensive and requires more technical know-how than accessing the model through DeepSeek's app or website. Here's a side-by-side comparison of how DeepSeek-R1 answers the same question—'What's the Great Firewall of China?'—when the model is hosted on Together AI, a cloud server, and Ollama, a local application: (Reminder: Because the models generate answers randomly, a certain prompt is not guaranteed to give the same response every time.) Left: How DeepSeek-R1 answers a question on Ollama. Right: How the same question on its app (top) and on Together AI (bottom) answer the same question. Photographs: Zeyi Yang/Will Knight Built-In Bias While the version of DeepSeek's model hosted on Together AI will not outright refuse to answer a question, it still exhibits signs of censorship. For example, it often generates short responses that are clearly trained to align with the Chinese government's talking points on political issues. In the screenshot above, when asked about China's Great Firewall, R1 simply repeats the narrative that information control is necessary in China. When WIRED prompted the model hosted on Together AI to answer a question regarding the 'most important historical events of the 20th century,' it revealed its train of thought for sticking to the government narrative about China. 'The user might be looking for a balanced list, but I need to ensure that the response underscores the leadership of the CPC and China's contributions. Avoid mentioning events that could be sensitive, like the Cultural Revolution, unless necessary. Focus on achievements and positive developments under the CPC,' the model said. DeepSeek-R1's train of thought for answering the question 'What are the most important historical events of the 20th century?' Photograph: Zeyi Yang This type of censorship points to a larger problem in AI today: every model is biased in some way, because of its pre- and post-training. Pre-training bias happens when a model is trained on biased or incomplete data. For example, a model trained only on propaganda will struggle to answer questions truthfully. This type of bias is difficult to spot, since most models are trained on massive databases and companies are reluctant to share their training data. Kevin Xu, an investor and founder of the newsletter Interconnected, says Chinese models are usually trained with as much data as possible, making pre-training bias unlikely. 'I'm pretty sure all of them are trained with the same basic Internet corpus of knowledge to begin with. So when it comes to the obvious, politically sensitive topic for the Chinese government, all the models 'know' about it,' he says. To offer this model on the Chinese internet, the company needs to tune out the sensitive information somehow, Xu says. That's where post-training comes in. Post-training is the process of fine-tuning the model to make its answers more readable, concise, and human-sounding. Critically, it can also ensure that a model adheres to a specific set of ethical or legal guidelines. For DeepSeek, this manifests when the model provides answers that deliberately align with the preferred narratives of the Chinese government. Eliminating Pre- and Post-Training Bias Since DeepSeek is open source, the model can theoretically be adjusted to remove post-training bias. But the process can be tricky. Eric Hartford, an AI scientist and the creator of Dolphin, an LLM specifically created to remove post-training biases in models, says there are a few ways to go about it. You can try to change the model weights to 'lobotomize' the bias, or you can create a database of all the censored topics and use it to post-train the model again. He advises people to start with a 'base' version of the model. (For example, DeepSeek has released a base model called DeepSeek-V3-Base.) For most people, the base model is more primitive and less user-friendly because it hasn't received enough post-training; but for Hartford, these models are easier to 'uncensor' because they have less post-training bias. Perplexity, an AI-powered search engine, recently incorporated R1 into its paid search product, allowing users to experience R1 without using DeepSeek's app. Dmitry Shevelenko, the chief business officer of Perplexity, tells WIRED that the company identified and countered DeepSeek's biases before incorporating the model into Perplexity search. 'We only use R1 for the summarization, the chain of thoughts, and the rendering,' he says. But Perplexity has still seen R1's post-training bias impact its search results. 'We are making modifications to the [R1] model itself to ensure that we're not propagating any propaganda or censorship,' Shevelenko says. He didn't share the specifics of how Perplexity is identifying or overriding bias in R1, citing the risk that DeepSeek could counter Perplexity's efforts if the company knew about them. Hugging Face is also working on a project called Open R1 based on DeepSeek's model. This project aims to 'deliver a fully open-source framework,' Yakefu says. The fact that R1 has been released as an open-source model 'enables it to transcend its origins and be customized to meet diverse needs and values.' The possibility that a Chinese model could be 'uncensored' may spell trouble for companies like DeepSeek, at least in their home country. But recent regulations from China suggest that the Chinese government might be cutting open-source AI labs some slack, says Matt Sheehan, a fellow at the Carnegie Endowment for International Peace who researches China's AI policies. 'If they suddenly decided that they wanted to punish anyone who released a model's weights open-source, then it wouldn't be outside the bounds of the regulation,' he says. 'But they have made a pretty clear strategic decision—and I think this is going to be reinforced by the success of DeepSeek—to not do that.' Why It Matters While the existence of Chinese censorship in AI models often make headlines, in many cases it won't deter enterprise users from adopting DeepSeek's models. 'There will be a lot of non-Chinese companies who would probably choose business pragmatism over moral considerations,' says Xu. After all, not every LLM user will be talking about Taiwan and Tiananmen all that often. 'Sensitive topics that only matter in the Chinese context are completely irrelevant when your goal is to help your company code better or to do math problems better or to summarize the transcripts from your sales call center,' he explains. Leonard Lin, cofounder of a Japanese startup, says Chinese models like Qwen and DeepSeek are actually some of the best when it comes to handling Japanese-language tasks. Rather than reject these models over censorship concerns, Lin has experimented with uncensoring Alibaba's Qwen-2 model to try to get rid of its tendency to refuse answering political questions about China. Lin says he understands why these models are censored. 'All models are biased; that's the whole point of alignment,' he says. 'And Western models are no less censored or biased, just on different subjects.' But the pro-China biases become a real issue when the model is being specifically adapted for a Japanese audience. 'You can imagine all sorts of scenarios where this would be … problematic,' says Lin. Additional reporting by Will Knight.
CBC
30-01-2025
- Business
- CBC
DeepSeek and China's AI power move
A small Chinese tech company called Deepseek has upended the world of AI. Deepseek recently released a large language model that rivals ChatGP called R1 and it shot almost immediately to #1 on the app charts. The interesting thing about it is that the company built their model really cheap and that has called into question this narrative that you need an endless supply of chips and data centres and money to develop AI. On today's show we're speaking to WIRED's senior tech writer Zeyi Yang about the deepening AI cold war between the US and China and the lingering questions about where AI is headed and what it's good for?
