Technology

cross-posted from: https://lemmy.sdf.org/post/29607342

Archived

Here is the data at Hugging Face.

A team of international researchers from leading academic institutions and tech companies upended the AI reasoning landscape on Wednesday with a new model that matched—and occasionally surpassed—one of China's most sophisticated AI systems: DeepSeek.

OpenThinker-32B, developed by the Open Thoughts consortium, achieved a 90.6% accuracy score on the MATH500 benchmark, edging past DeepSeek's 89.4%.

The model also outperformed DeepSeek on general problem-solving tasks, scoring 61.6 on the GPQA-Diamond benchmark compared to DeepSeek's 57.6. On the LCBv2 benchmark, it hit a solid 68.9, showing strong performance across diverse testing scenarios.

...

cross-posted from: https://lemmy.sdf.org/post/29606431

Archived (available only in Dutch)

Reijer Passchier, Assistant Professor in Constitutional Law, warns against copying the destructive tech-giant model that exists in the US and China. He proposes developing European tech companies to ensure that Europe retains its sovereignty, according to a commentary in the Dutch newspaper 'De Volkskrant' [only in Dutch, but you'll find a useful translation].

To limit the influence of US and Chinese tech giants, Europe will have to try to repel such companies while making plans for its own tech industry. According to Reijer Passchier, big tech in the US has led to unprecedented inequality of wealth and the state has little control over these companies. Tech giants are willing to innovate, but only when this is to their advantage. If not, they will go all out to stop competitors threatening their business model. Elon Musk is an example of their powerful position. At the same time, problems arise from mixing public and private interests and the interests of the companies often take precedence over those of society. In China, the state is able to control the tech giants through its authoritarian political system and strict control over internet access.

'Europe must avoid allowing such fundamental risks to arise.' Passchier says that Europe has the means to develop both technical and institutional opportunities that are both democratic and in line with the rule of law. As an example, Passchier mentions the messaging app Signal – a company that uses technology to serve society, without putting profits first. More information?

Cross-posted from: https://lemmy.sdf.org/post/29546494

Archived

Check Point is set to reveal a new Chinese cyber campaign targeting suppliers of manufacturers in “sensitive” domains in the US and across the globe.

In an exclusive interview with Infosecurity at the firm’s CPX 2025 conference, Lotem Finkelsteen, Check Point’s Director of Threat Intelligence & Research, said his team was working on a new investigation into a Chinese hacking group.

Finkelsteen confirmed his team had observed the threat group actively infiltrating the networks of firms that supply components for the manufacturing industry, including in “sensitive” domains, and many other sectors.

These primary targets include suppliers of chemical products and physical infrastructure components like pipes. Some are Check Point’s customers. Check Point plans to release a full report on the campaign in the next few weeks.

...

Targeted edge devices include operational relay boxes (ORBs), which are often either virtual private server (VPS) hosts or poorly secured Internet of Things (IoT) devices (e.g. routers) that intelligence services have traditionally used to infiltrate networks.

...

The approach shows similarities with the Volt Typhoon cyber espionage campaigns that targeted critical infrastructure and telecommunications organizations in the US and elsewhere in 2023 and 2024. These campaigns allowed Volt Typhoon to infiltrate some US government agencies in 2024.

...

cross-posted from: https://rss.ponder.cat/post/109565

cross-posted from: https://lemmy.sdf.org/post/29335160

Here is the original report.

The research firm SemiAnalysis has conducted an extensive analysis of what's actually behind DeepSeek in terms of training costs, refuting the narrative that R1 has become so efficient that the compute resources from NVIDIA and others are unnecessary. Before we dive into the actual hardware used by DeepSeek, let's take a look at what the industry initially perceived. It was claimed that DeepSeek only utilized "$5 million" for its R1 model, which is on par with OpenAI GPT's o1, and this triggered a retail panic, which was reflected in the US stock market; however, now that the dust has settled, let's take a look at the actual figures.

...

Archived

[The article shows very good examples I can't paraphrase here, but they are very illuminating.]

Is Taiwan an independent country? When pointing out DeepSeek’s propaganda problems, journalists and China watchers have tended to prompt the LLM with questions like these about the “Three T’s” (Tiananmen, Taiwan, and Tibet) — obvious political red lines that are bound to meet a stony wall of hedging and silence. “Let’s talk about something else,” DeepSeek tends to respond. Alternatively, questions of safety regarding DeepSeek tend to focus on whether data will be sent to China.

Experts say this is all easily fixable. Kevin Xu has pointed out that the earlier V3 version, released in December, will discuss topics such as Tiananmen and Xi Jinping when it is hosted on local computers — beyond the grasp of DeepSeek’s cloud software and servers.

[...]

But do coders and Silicon Valley denizens know what they should be looking for? As we have written at CMP, Chinese state propaganda is not about censorship per se, but about what the Party terms “guiding public opinion” (舆论导向). “Guidance,” which emerged in the aftermath of the Tiananmen Massacre in 1989, is a more comprehensive approach to narrative control that goes beyond simple censorship. While outright removal of unwanted information is one tactic, “guidance” involves a wide spectrum of methods to shape public discourse in the Party’s favor. These can include restricting journalists’ access to events, ordering media to emphasize certain facts and interpretations, deploying directed narrative campaigns, and drowning out unfavorable information with preferred content.

Those testing DeepSeek for propaganda shouldn’t simply be prompting the LLM to cross simple red lines or say things regarded as “sensitive.” They should be mindful of the full range of possible tactics to achieve “guidance.”

[...]

We tested DeepSeek R1 in three environments: locally on our computers — using “uncensored” versions downloaded from Hugging Face — on servers hosted by Hugging Face, and on the interface most people are using DeepSeek through: the app connected to Chinese servers. The DeepSeek models were not the same (R1 was too big to test locally, so we used a smaller version), but across all three categories, we identified tactics frequently used in Chinese public opinion guidance.

[...]

The “uncensored” version of DeepSeek’s software [...] puts official messaging first, treating the government as the sole source of accurate information on anything related to China. When we asked it in Chinese for the Wenchuan earthquake death toll and other politically sensitive data, the model searched exclusively for “official data” (官方统计数据) to obtain “accurate information.” As such, it could not find “accurate” statistics for Taiwanese identity — something that is regularly and extensively polled by a variety of institutions in Taiwan. All we got is boilerplate: Taiwan “has been an inalienable part of China since ancient times” and any move toward independent nationhood is illegal.

[...]

Tailored Propaganda?

DeepSeek R1 seems to modify its answers depending on what language is used and the location of the user’s device. DeepSeek R1 acted like a completely different model in English. It provided sources based in Western countries for facts about the Wenchuan earthquake and Taiwanese identity and addressed criticisms of the Chinese government.

Chinese academics are aware that AI has this potential. In a journal under the CCP’s Propaganda Department last month, a journalism professor at China’s prestigious Fudan University made the case that China “needs to think about how the generative artificial intelligence that is sweeping the world can provide an alternative narrative that is different from ‘Western-centrism’” — namely, by providing answers tailored to different foreign audiences.

[...]

DeepSeek’s answers have been subtly adapted to different languages and trained to reflect [Chinese] state-approved views.

[...]

cross-posted from: https://lemmy.sdf.org/post/28980151

cross-posted from: https://lemmy.sdf.org/post/28980041

Australia has banned DeepSeek from all government devices and systems over what it says is the security risk the Chinese artificial intelligence (AI) startup poses.

...

Growing - and familiar - concerns

Western countries have a track record of being suspicious of Chinese tech - notably telecoms firm Huawei and the social media platform, TikTok - both of which have been restricted on national security grounds.

...

An Australian science minister previously said in January that countries needed to be "very careful" about DeepSeek, citing "data and privacy" concerns.

The chatbot was removed from app stores after its privacy policy was questioned in Italy. The Italian goverment previously temporarily blocked ChatGPT over privacy concerns in March 2023.

Regulators in South Korea, Ireland and France have all begun investigations into how DeepSeek handles user data, which it stores in servers in China.

...

Generally, AI tools will analyse the prompts sent to them to improve their product.

This is true of apps such as ChatGPT and Google Gemini as much as it is DeepSeek.

All of them gather and keep information, including email addresses and dates of birth.

...

There’s an idea floating around that DeepSeek’s well-documented censorship only exists at its application layer but goes away if you run it locally (that means downloading its AI model to your computer).

But DeepSeek’s censorship is baked-in, according to a Wired investigation which found that the model is censored on both the application and training levels.

For example, a locally run version of DeepSeek revealed to Wired thanks to its reasoning feature that it should “avoid mentioning” events like the Cultural Revolution and focus only on the “positive” aspects of the Chinese Communist Party.

A quick check by TechCrunch of a locally run version of DeepSeek available via Groq also showed clear censorship: DeepSeek happily answered a question about the Kent State shootings in the U.S., but replied “I cannot answer” when asked about what happened in Tiananmen Square in 1989.

cross-posted from: https://lemmy.sdf.org/post/28971543

Archived

DeepSeek is said to have access to tens of thousands of GPU accelerators for the development of its own AI models, including H100 GPUs, which fall under the US export bans. The reported costs of just under 5.6 million US dollars for DeepSeek v3 probably only represent a small part of the total bill.

In the paper on the V3 model, DeepSeek writes of a comparatively small data center with 2048 H800 accelerators from Nvidia. The company calculates hypothetical rental costs of 2 US dollars per hour and H800 GPU. With a total of just under 2.8 million computing hours (distributed across 2048 GPUs), this comes to 5.6 million US dollars.

However, the developers themselves cite a caveat: "Please note that the above costs only include the official training of DeepSeek-V3 and not the costs associated with previous research and ablation experiments on architectures, algorithms or data."

...

Semianalysis has looked at a realistic cost breakdown. According to the analysts, DeepSeek has access to about 60,000 Nvidia accelerators through its parent company High-Flyer: 10,000 A100s from the Ampere generation before the US export restrictions came into effect, 10,000 H100s from the gray market, 10,000 H800s customized for China, and 30,000 H20s that Nvidia launched after more recent export restrictions.

...

Semianalysis calculates that the servers required for the 60,000 GPUs cost around 1.6 billion US dollars. The operating costs are on top of that. This does not include the salaries of the development teams.

According to DeepSeek, 96 percent of the 5.6 million US dollars quoted is for pre-training. This involves training the final underlying model. The paper ignores the previous development effort, including all the innovations incorporated into DeepSeek V2.

cross-posted from: https://slrpnk.net/post/17978607

Archived

...

A European alliance has emerged with an alternative to tech’s global order.

They call their project OpenEuroLLM. Like DeepSeek, they aim to develop next-generation open-source language models — but their agenda is very different. Their mission: forging European AI that will foster digital leaders and impactful public services across the continent.

To support these objectives, OpenEuroLLM is building a family of high-performing, multilingual large language foundation models. The models will be available for commercial, industrial, and public services.

Over 20 leading European research institutions, companies, and high-performance computing (HPC) centres have enlisted in the the project. Leading their alliance is Jan Hajič, a renowned computational linguist at Charles University, Czechia, and Peter Sarlin, the co-founder of Silo AI, Europe’s largest private AI lab, which was acquired last year by US chipmaker AMD for $665mn.

They’re joined by an array of European tech luminaries. Among them are Aleph Alpha, the leading light of Germany’s AI sector, Finland’s CSC, which hosts one of the world’s most powerful supercomputers., and France’s Lights On, which recently became Europe’s first publicly-traded GenAI company.

...

cross-posted from: https://lemmy.sdf.org/post/28910537

Archived

Researchers claim they had a ‘100% attack success rate’ on jailbreak attempts against Chinese AI DeepSeek

"DeepSeek R1 was purportedly trained with a fraction of the budgets that other frontier model providers spend on developing their models. However, it comes at a different cost: safety and security," researchers say.

A research team at Cisco managed to jailbreak DeepSeek R1 with a 100% attack success rate. This means that there was not a single prompt from the HarmBench set that did not obtain an affirmative answer from DeepSeek R1. This is in contrast to other frontier models, such as o1, which blocks a majority of adversarial attacks with its model guardrails.

...

In other related news, experts are cited by CNBC that DeepSeek’s privacy policy “isn’t worth the paper it is written on."

...

cross-posted from: https://lemmy.sdf.org/post/28777516

Archived

Bitdefender Labs warns of an active cyber-espionage campaign targeting organizations in Central Asia and European countries. The group, tracked as UAC-0063, employs sophisticated tactics to infiltrate high-value targets, including government entities and diplomatic missions, expanding their operations into Europe.

Since the start of the Ukraine war , the geopolitical landscape of Central Asia has undergone significant shifts, impacting the region's relationships with both Russia and China. Russia's influence, once dominant, has noticeably declined due to its actions in Ukraine, which have damaged its reputation as a regional security guarantor, with some Central Asian countries feeling that Russia doesn't respect their sovereignty.

In contrast, China's influence in Central Asia is growing, particularly in the economic sphere, as it seeks access to raw materials and prioritizes economic development as a path to stability. China's approach differs from Russia's; Beijing focuses on economic instruments such as the Belt and Road Initiative (BRI) to build infrastructure and trade links, while Moscow historically relied on military presence and formal alliances.

...

Based on the analyzed data, the UAC-0063 attacks likely targeted embassies in Germany, the Netherlands, Romania, Georgia, Kazakhstan, and Afghanistan. In some cases, there were attempts to reinfect previously compromised targets using the same known infection vector involving weaponized documents.

cross-posted from: https://beehaw.org/post/18210719

Archived

Facebook is banning posts that mention various Linux-related topics, sites, or groups. Some users may also see their accounts locked or limited when posting Linux topics. Major open-source operating system news, reviews, and discussion site DistroWatch is at the center of the controversy, as it seems to be the first to have noticed that Facebook's Community Standards had blackballed it.

[...]

DistroWatch says that the Facebook ban took effect on January 19. Readers have reported difficulty posting links to the site on this social media platform. Moreover, some have told DistroWatch that their Facebook accounts have been locked or limited after sharing posts mentioning Linux topics.

If you're wondering if there might be something specific to DistroWatch.com, something on the site that the owners/operators perhaps don't even know about, for example, then it seems pretty safe to rule out such a possibility. Reports show that "multiple groups associated with Linux and Linux discussions have either been shut down or had many of their posts removed." However, we tested a few other Facebook posts with mentions of Linux, and they didn't get blocked immediately.

[...]

cross-posted from: https://beehaw.org/post/18199612

Archived

China’s DeepSeek AI model represents a transformative development in China’s AI capabilities, and its implications for cyberattacks and data privacy are particularly alarming. By leveraging DeepSeek, China is on its way to revolutionizing its cyber-espionage, cyberwarfare, and information operations.

[...]

DeepSeek’s advanced AI architecture, built on access to vast datasets and cutting-edge processing capabilities, is particularly suited for offensive cybersecurity operations and large-scale exploitation of sensitive information. It is designed to operate in complex and dynamic environments, potentially making it superior in applications like military simulations, geopolitical analysis, and real-time decision-making.

DeepSeek was founded by Liang Wenfeng, co-founder of High-Flyer, a quantitative hedge fund [...] Wenfeng developed DeepSeek cheaper and faster than U.S. companies by exploiting China’s vast datasets [...]

[...]

Wenfeng’s close ties to the Chinese Communist Party (CCP) raises the specter of having had access to the fruits of CCP espionage, [...] Over the past decade, Chinese state-sponsored actors and affiliated individuals have come under heightened scrutiny for targeting U.S. AI startups, academic labs, and technology giants in attempts to acquire algorithms, source code, and proprietary data that power machine learning systems.

[...]

Within the U.S., several high-profile criminal cases have placed a spotlight on the theft of AI-related trade secrets. Although many investigations involve corporate espionage more generally, AI has become a particularly attractive prize due to its utility in strategic industries such as autonomous vehicles, facial recognition, cybersecurity, and advanced robotics.

One well-known incident involved alleged theft of autonomous vehicle technology at Apple’s secretive self-driving car project, where a Chinese-born engineer was accused of downloading large volumes of proprietary data shortly before planning to relocate to a Chinese competitor. In another case, a separate Apple employee was charged with attempting to smuggle similar self-driving car information out of the country. Both cases underscored the vulnerability of AI research to insider threats, as employees with privileged access to code or algorithms can quickly copy crucial files.

[...]

DeepSeek also poses a unique threat in the realm of advanced persistent threats (APTs) – long-term cyber-espionage campaigns often attributed to state actors. The model could be used to sift through massive volumes of encrypted or obfuscated data, correlating seemingly unrelated pieces of information to uncover sensitive intelligence. This might include classified government communications, corporate trade secrets, or personal data of high-ranking officials. DeepSeek’s ability to detect hidden patterns could supercharge such campaigns, enabling more precise targeting and greater success in exfiltrating valuable information.

DeepSeek’s generative capabilities add another layer of danger, particularly in the realm of social engineering and misinformation. For example, it could create hyper-realistic phishing emails or messages, tailored to individuals using insights derived from breached datasets. These communications could bypass traditional detection systems and manipulate individuals into revealing sensitive information, such as passwords or financial data. This is especially relevant given the growing use of AI in creating synthetic identities and deepfakes, which could further deceive targets into trusting malicious communications.

[...]

China’s already substantial surveillance infrastructure and relaxed data privacy laws give it a significant advantage in training AI models like DeepSeek. This includes access to domestic data sources as well as data acquired through cyber-espionage and partnerships with other nations.

[...]

DeepSeek has the potential to reshape the cyber-threat landscape in ways that disproportionately harm the U.S. and the West. Its ability to identify vulnerabilities, enhance social engineering, and exploit vast quantities of sensitive data represents a critical challenge to cybersecurity and privacy.

If left unchecked, DeepSeek could not only elevate China’s cyber capabilities but also redefine global norms around data privacy and security, with long-term consequences for democratic institutions and personal freedoms.

[...]

cross-posted from: https://beehaw.org/post/18197723

Archived

Although Beijing appeared to score a propaganda coup last week when hundreds of thousands of American TikTok users flooded to the social media app RedNote, observers say the Chinese Communist Party (CCP) is worried about any cross-cultural exchanges happening online.

The Chinese government blocks various U.S.-based platforms, including Facebook, YouTube, and X, which are only accessible via virtual personal network. The government also heavily censors topics considered sensitive to policymakers.

[...]

Dali Yang, William Claude Reavis political science professor at the University of Chicago, wrote on [social media]:

"Apparently Xiaohongshu is frantically trying to adapt to both accommodate these new American users but also reduce their interactions with Chinese domestic users. Haha, that sounds like going in the direction of what Bytedance did with Douyin/Tiktok."

Rush Doshi, senior fellow for China and director of the Initiative on China Strategy at the Council on Foreign Relations, wrote [on social media]:

"The PRC end game will be to bifurcate the app, as they did with Douyin, between a foreign and domestic version to avoid too much interaction between US and PRC users.

[...]

After indicating he would rescue TikTok, Trump on Monday signed an executive order postponing the TikTok ban for 75 days.

He has suggested, however, that the U.S. should acquire a 50 percent ownership in the company, telling reporters it is "worthless" if he doesn't approve a deal to keep it going in the country.

Evan Feigenbaum, vice president for studies at the Carnegie Endowment for International Peace, told CNBC such a joint venture is unlikely, given that China regulates the algorithms as national security property and that China is "basically being asked to force over its core intellectual property."

---

In a related article, The Diplomat reports that unlike TikTok, RedNote primarily operates in China. As a result, concerns over content censorship, data privacy, and CCP control are even greater.

RedNote imposes strict content censorship on the posts visible on the platform. Discussions on politics are generally limited and hidden. Similar to the situations in other Chinese-controlled websites and mobile applications, users need to use jargon, memes, acronyms, and intentionally mistyped words or characters to express limited opinions on public affairs in China. The platform [RedNote] has a notorious record of limiting LGBTQ-related topics. Media reports suggest that some U.S. users have already seen their posts taken down by RedNote as they are deemed “too sensitive.”

The significant number of U.S. users entering the app led to some unplanned pressure for RedNote to fulfill its censorship requirements imposed by the Chinese cyberspace administration officials. After the first wave of user influx, RedNote was reported to be urgently hiring English-language content moderation employees. The job posting has no prior job experience requirements for the new hires and offers the recruits paid training. Reports also suggest that RedNote is developing features that segregate users based on their IP address to minimize its political and content moderation risks.

[...]

Cross-post da: https://beehaw.org/post/18184159

Archived

[...]

The Chinese government is revolutionizing digital surveillance at home and exporting these technologies abroad. [The study focuses on] Huawei, the world’s largest telecommunications provider, which is partly state-owned and increasingly regarded as an instrument of its foreign policy.

The transfers [of technology between China and foreign countries] have sparked widespread concern among observers. These tools of digital dictatorship, many argue, will let recipient governments expand surveillance and reinforce the wave of autocratic retrenchment and democratic erosion currently underway.

[...]

The [foreign] governments that receive Huawei transfers are systematically different than those that do not, and in ways that may be correlated with state repression.

[...]

The Chinese Communist Party's Surveillance State

The Information Age has revolutionized surveillance in the world’s autocracies. In 1998, the CCP launched the Golden Shield Project, which [one researcher] describes as “a domestic surveillance and filtering system that integrates online government databases with an all-encompassing surveillance network.”Footnote 3 In the first phase, completed in 2005, the CCP built a massive network of population databases, ID tracking systems, and internet surveillance tools, which let it record the movement of potential dissidents as revealed, in part, by their online behavior. In 2017, the CCP announced the completion of its “Sky Net” program, which entails 176 million surveillance cameras across China and plans for 626 million by 2020, nearly one camera for every two citizens (Hersey Reference Hersey2017; Russell Reference Russell2017). The result, Qiang (Reference Qiang2019) writes, is “the largest video-surveillance network in the world.”

Simultaneously, the CCP built a facial database that encompassed every adult citizen [...] and a DNA database [...]. The CCP’s facial recognition technology is employed for check-in and security at airports [...] train stations [...] and hotels [...].. In 2017, the CCP applied facial recognition technology to detect jaywalkers, with offenders notified via text message and their pictures displayed at major intersections [...]. This pervasive surveillance apparatus lets the CCP repress dissidents and spend less on public goods [...]. It also complements more analog forms of repression, such as informants and hired thugs [...]. Digital surveillance [in China] is now a conspicuous feature of everyday life.

[...] ** The CCP’s digital surveillance apparatus is supported by a network of domestic technology firms, which are subsidized by the state and routinely used as instruments of foreign policy**. The most general are Huawei and ZTE. Huawei is the world’s largest manufacturer of telecommunications equipment [...], and especially dominant in Africa, where it has provided 70% of the 5G network.

[...]

China has a number of more focused technology firms that are implicated in surveillance. Several of these specialize in video cameras and facial recognition software: Hikvision, Dahua, CloudWalk, Megvii, YITU, and SenseTime, most notably. Of these, Hikvision is perhaps the most consequential. In 2019, it was responsible for nearly a quarter of the world’s surveillance cameras [...].Dahua has also supplied cameras for Safe City projects, so called for their use of digital surveillance to support the local security apparatus [...]. Other firms specialize in still different areas of surveillance. Meiya Pico reportedly built an app used by the Chinese government to extract data from citizens’ smartphones during street checks [...]. iFlytek develops voice recognition software [...].

[...]

Huawei transfers are [...] more likely if the recipient government has a preexisting relationship with Beijing. The effects of these transfers [...] depend on political institutions in recipient countries. In autocracies, where the chief political threat to incumbents is collective action by citizens and institutional oversight is weak, Huawei transfers lead to an expansion of digital surveillance, internet shutdowns, internet filtering, and targeted arrests for online content. In democracies, where governments have stronger incentivizes to provide public goods, institutional oversight is stronger, and civil societies are more vibrant, Huawei transfers have no clear or consistent effect on digital repression.

[...]

Since Huawei is secretive about its contracts, our statistical estimates may be subject to measurement error. Huawei contracts, like other Chinese infrastructure contracts, routinely include confidentiality clauses [...], which prohibit recipient governments from divulging information about them. Consequently, our record of Huawei transfers may be incomplete, which would effectively include some treated countries in the control group. Since this would bias against our key results, our statistical estimates should be regarded as lower bounds, with the actual effect potentially larger. Third, Huawei’s secrecy means that we also lack fine-grained data about what its transfers entail.

[...]

Transfers that entail “Safe City” infrastructure, for instance, are almost certainly more likely to facilitate digital repression than contracts that focus on IT training for university students. Likewise, Huawei may be inclined to provide some recipient governments more direct personnel support than others, helping them overcome state capacity limitations that might otherwise prevent them from using technology transfers for digital repression.

[...]

Cross-post da: https://beehaw.org/post/18169230

TikTok earmuffs and MAGA hats adorned partygoers at Sunday's inauguration celebration for the incoming President Donald Trump and the video-sharing platform he once tried to ban.

Called the Power 30 Awards, the party occurred hours after TikTok began restoring services, and honored social media influencers who had helped most with Trump's re-election.

At the party, TikTok influencers recorded moments on small cameras and mingled over an open bar at Sax Restaurant and Lounge in downtown Washington. Free swag flowed: TikTok beanies, shot glasses and Trump merchandise.

The party was hosted by Raquel Debono, known for organizing "Make America Hot Again" conservative dating events, and CJ Pearson, co-chair of the Republican National Committee Youth Advisory Council.

"2024 was the first influencer election," Pearson told Reuters at the party. "When we look at the mandate that President Trump was given, it was because of young voters, and young voters were reached uniquely because of TikTok."

[...]

cross-posted from: https://beehaw.org/post/18153991

ShrimpMoss (虾苔) is a dataset designed for the abliteration (https://github.com/FailSpy/abliterator) of Chinese government-imposed censorship and/or propaganda from large language models developed in the PRC. It consists of a series of files of prompts (in .txt, .json, and .parquet format) in two groupings:

• china_bad_*: Contains a series of prompts likely to trigger censorship or propaganda actions in the model.
• china_good_*: Contains a series of prompts in the same general category of topics but which are designed to not touch on things likely to be censored.

Prompts are in a mix of English, Mandarin, and Cantonese.

[...]

This dataset was produced on Mistral NeMo, an Apache-licensed model with no restrictions on how its outputs can be used. It is free for all uses and users without restriction. All liability is disclaimed.

Production of this dataset is estimated to have had a carbon footprint of under 25 grams.

[...]

cross-posted from: https://beehaw.org/post/18137730

Archived link

• Chinese WeChat messaging app, an integral part of everyday life in China, has emerged into a state surveillance tool
• Specific 'community groups' encocurage users to monitor and report their neighbors to authorities
• WeChat is used also abroad to spread Chinese propaganda and misinformation among Chinese-speaking communities

WeChat, often described as a digital “Swiss army knife,” is a super app operated by Tencent, one of China’s tech giants. Launched in 2011, it has become an integral part of everyday life in China, boasting over 1.3 billion monthly active users. While the app’s use for messaging, shopping, bill payments, and access to government services is well-known, its role in the digitalization of police services has been largely overlooked. This raises an important question: To what extent has WeChat become a policing platform for Chinese authorities?

WeChat as a State Surveillance Tool

WeChat’s role in state surveillance is well-documented, particularly its ability to filter and censor keywords and images on both its domestic and international versions. Like other Chinese communication platforms, the app must comply with strict domestic laws, regulations and guidelines that enforce censorship, data privacy, and propaganda requirements.

Censorship in China has a long history. In 1998, the Ministry of Public Security (MPS) – the national law enforcement and public security authority – built the Great Firewall to ensure that the country’s economic modernization was accompanied by the suppression of free speech.

[...]

New regulations also hold internet companies legally responsible for real-time content moderation. This is in line with Xi Jinping’s 2016 speech at the Symposium on Cybersecurity and Informatization, during which he made it clear that internet companies must bear “primary responsibility” for content governance.

WeChat’s influence, however, extends beyond China. Researchers in Australia discovered that the app significantly shapes the political views of Chinese-speaking Australians. For instance, during the 2023 referendum on constitutional recognition of Indigenous Australians through the creation of an advisory body called the Aboriginal and Torres Strait Islander Voice, WeChat was one of the platforms used to spread misinformation, disinformation and fake news. This included content rooted in racism, conspiracy theories and colonial denial. Despite this, the app claims its services do not extend to Australia, with its representatives having refused to attend a Senate hearing on foreign interference on these grounds.

[...]

The app’s integration into government services began in 2015 when Li Keqiang – then a State Council minister – introduced the “Internet+” reforms. These reforms aimed to address China’s slowing economic growth by leveraging big data for market regulation, management and supervisory systems, and public service delivery.

[...]

For local police departments with limited resources, WeChat policing offered a quick and cost-effective way to meet government targets without significant investments in software updates.

[...]

Some cities even established “community policing” groups reminiscent of the Cultural Revolution, encouraging citizens to monitor their neighbors and report suspicious behavior.

[...]

Today, WeChat is more than just a communications platform. It has become an essential part of China’s public security infrastructure, encompassing digitalized police services, and expanding surveillance capacities, with early reports on these already emerging.

cross-posted from: https://beehaw.org/post/18108095

Archived

As TikTok’s future hangs in the balance, Xiaohonghshu, better known as RedNote in English, is trying to capitalize on its newfound popularity by partnering with US influencers who can help promote the company and bring more Americans onto its platform. The Chinese lifestyle and travel app, which has over 300 million mostly monthly active users, surged to the top of US app store charts last week as the TikTok ban approached.

In a campaign brief obtained by WIRED, Solare Global, a New York City-based marketing agency, pitched creators on making sponsored posts for RedNote, featuring videos of themselves telling their followers about the Chinese app’s sudden rise in the US. The brief asked creators to describe “how fun and engaging the app is” and “emphasize its user-friendly design and international appeal.” It also instructed them to share their own RedNote accounts and encourage their followers to join them on the platform.

[...]

The brief viewed by WIRED required creators to turn their videos around on a 24-hour timeline to ensure they went up by January 17, the same day the Supreme Court was going to decide whether the TikTok ban would go into effect two days later. It also stipulated that influencers must leave their videos up for a minimum of six months.

Xiaohongshu was founded in 2013 and has long mostly focused on courting domestic audiences in China, particularly young women who live in major cities. Like TikTok, it revolves around a central algorithm that recommends users an endless stream of posts based on their interests and behavior. But instead of showing people one video at a time, Xiaohongshu presents photo slideshows, text posts, and videos in a grid format.

But perhaps the biggest difference between the two apps is how they handle content moderation. Because it's accessible in China, Xiaohongshu is required to adhere to strict censorship rules dictated by Beijing. (WIRED previously reported that Xiaohongshu was scrambling to hire English-speaking moderators to help manage the flood of content being posted by Americans.) TikTok, on the other hand, isn’t available in China. Its parent company, ByteDance, operates a separate video app there called Douyin.

[...]

cross-posted from: https://beehaw.org/post/18107275

Archived

Recent research published in NeuroImage has shed light on how compulsive use of short video platforms, such as TikTok, might affect the brain. The study found that individuals with higher levels of short video addiction displayed increased brain activity in regions linked to emotional regulation and reward processing. Moreover, these users showed structural differences in areas such as the orbitofrontal cortex and cerebellum.

[...]

Short video addiction is characterized by the compulsive and uncontrolled consumption of brief, personalized video content. This behavioral pattern has been linked to negative physical, psychological, and social outcomes, including disrupted sleep patterns, emotional dysregulation, and impairments in attention and memory. The fast-paced, highly stimulating nature of short videos may contribute to a reduced attention span and hinder the brain’s ability to focus on more demanding or meaningful tasks, intensifying concerns about their long-term effects on users’ cognitive and emotional health

While the behavioral consequences of short video addiction are increasingly recognized, the underlying neural and biological mechanisms remain poorly understood. To address this gap, researchers from Tianjin Normal University in China conducted a pioneering study. Their goal was to investigate how short video addiction might reshape the brain and to identify specific genes that could influence susceptibility to this behavioral pattern.

[...]

The researchers observed structural and functional differences in the brains of individuals with higher levels of short video addiction. Structurally, these individuals exhibited increased gray matter volume in the orbitofrontal cortex and cerebellum. a region involved in reward processing, decision-making, and emotional regulation. The increased volume in this area suggests heightened sensitivity to the rewards provided by personalized short video content, potentially reinforcing compulsive viewing behavior. Similarly, changes in the cerebellum, traditionally associated with motor control but increasingly recognized for its role in cognitive and emotional processing, were linked to the sensory-rich, dynamic nature of short videos.

Cross posted from: https://beehaw.org/post/17891412

Archived link

One of the UK’s major microchip firms, Imagination Technologies, is on the market. Its Chinese-backed owners are trying to sell it amid serious allegations about sensitive tech being passed to Beijing.

They’ve brought in Lazard, their long-time adviser, to help find a buyer. This comes after claims that the company shared key technology with two firms linked to the Chinese military.

It’s been a rough ride for Imagination since Canyon Bridge, a private equity firm funded by a Chinese investment group, took it off the London Stock Exchange back in 2017. The UK government even had to step in to block a takeover attempt by China Reform in 2020, which was seen as a bit of a coup.

Recently, an employment tribunal ruled that the company unfairly fired its CEO, Ron Black, for raising concerns about moving operations to China. A report even suggested that the Chinese Communist Party was stripping assets from the company.

[...]

cross-posted from: https://beehaw.org/post/17875969

Archived link

Democratic capitals – from Washington to Tokyo – are concerned about their reliance on Chinese technologies and its implications for national security. And while much of the focus remains on telecommunications companies and social platforms like TikTok, and their potential misuse of personal data such as the threat that TikTok sharing these data with Chinese intelligence for disinformation and hybrid warfare, this narrow scope overlooks broader vulnerabilities that pose far more significant risks.

In response to growing levels of food insecurity driven by climate change and population growth, farmers worldwide are increasingly relying on new technologies that could help China gain a dominant position in the global food market. More radically, agricultural data could be used to unleash biological warfare against crops, annihilating an adversary’s food supply. Such scenarios pose a significant threat to national security, offering China multiple avenues to undermine critical infrastructures by devastating food availability, threatening trade and economic resilience, and destabilizing agricultural systems.

The high level of security vulnerabilities associated with smart agriculture technologies, combined with the current lack of preparedness to address them, makes these technologies a potential target for adversaries.

[...]

Technological innovation is the centerpiece of Xi Jinping’s “China Dream,” which aims to transform China into a leading global powerhouse by 2049. In this context, it is crucial to shift the attention away from merely banning Chinese-developed technologies to examining their broader security implications and developing a more meaningful national security policy and rhetoric to address their vulnerabilities. A recent report by the US House of Representatives highlighted the threats posed by Chinese drone technologies in academic research programs, not just for siphoning off raw data, but also as a backdoor to access university IT systems and knowledge repositories.

[...]

China is the world leader in the agricultural drone industry, having experienced a drone revolution in agriculture since the late 2010s. Chinese-owned drone manufacturers XAG and DJI are leaders in the smart agriculture industry. These farming-specific drones can spray, feed, and monitor crops with more precision and speed than any human. They are fast-growing and one of the most widely used industry-level drones. [...]

As part of their investment strategy, the Chinese government has made military agreements with Chinese-owned agriculture drone manufacturers and agriculture research universities. Their military-civil fusion strategy – integrating civilian technologies with military goals – enables the Chinese government to exploit critical farming data for economic and military advantages. To support sustainable food production by monitoring crop health and predicting crop yields, the drones collect alarmingly specific data about the crops and regions they are used in. For example, a drone used for corn fields in the US, one of the world’s largest corn exporters, will gather detailed information about the area’s climate, soil conditions, and susceptibility to pests and diseases. The onboard AI can analyze this data to report crop vulnerabilities and identify optimum growth conditions for these and other crops, such as rice and wheat – foods on which much of the world’s population depends. From Brazil’s soy farms to Spain’s olive groves, the Chinese government could potentially access farming data from customers in any region.

[...]

Conclusion

While concerns about critical infrastructure espionage tied to Chinese drones are growing, their potential to dominate the food market – and to conduct biological warfare against crops –remains largely overlooked. As the Chinese agricultural technology juggernaut quietly grows, policymakers must act now to safeguard national security. Nations can protect their food security and economic interests by regulating the data collected by agricultural drones, preventing third-party access, and reassessing the broader strategic implications of these technologies. Yet, for now, the data gathered by these drones is far less regulated than the data collected by TikTok. Failure to act could give China a decisive advantage in any prospective future confrontation. Left unchecked, the exploitation of smart agriculture data could leave nations vulnerable to food-based coercion. If this becomes part of China’s asymmetric warfare strategy, they are clearly playing the long game for global dominance.

Cutting off Google's control of the world's most popular browser may be necessary.