PawsUp

1 readers
0 users here now
founded 2 months ago
ADMINS
1626
 
 
1627
 
 
1628
1629
1630
1631
 
 
1632
 
 
1633
 
 
1634
 
 
1635
1636
1637
1638
1639
1640
173
submitted 1 month ago* (last edited 1 month ago) by stumu415@lemmy.zip to c/technology@lemmy.world
1641
1642
 
 

It would be interesting to read the post mortem (if they will release). In particular in which phase of the training it escaped the sandbox, in which way it used the information. Training is different from inference, I doubt there is a context there and where would the model store these information? In the weights?! That makes little sense to me

1643
1644
1645
 
 

cross-posted from: https://lemmy.sdf.org/post/56505579

Here is the report in Polish (pdf) and the report in English (pdf).

How does TikTok show China to Polish users?

Experts from the Polish NASK Disinformation Analysis Centre examined nearly 10,000 pieces of material on six topics considered particularly sensitive by Beijing (Xinjiang, Tibet, Tiananmen Square massacre, Taiwan, human rights, re-education camps). They show that the way content is recommended may be conducive to building a favourable image of China.

Some key points:

  • Many of the materials classified as „pro-China” based on the adopted criteria contained statements fully consistent with messages approved by the Publicity Department and other bodies directly subordinate to the Chinese Communist Party (CCP).

  • There are at least several English-language accounts on the Tiktok app responsible for spreading pro-Chinese propaganda messages. A few characteristics indicate their direct links to the Chinese propaganda mechanism, including they exclusively discuss topics important to Chinese propaganda message, promote an extremely pro-Chinese message, attempt to discredit individuals criticizing the CCP’s actions, and publish large amounts of professionally prepared content despite posing as, for example, ordinary tourists in China.

  • The Tiktok app does not designate state-owned channels as “state-controlled media as Facebook does. This means that it is impossible to determine which channels on the platform operate directly under government control.

  • The examined material raises legitimate doubts about the impartiality of the TikTok app. It is impossible to definitively answer the question whether this is solely the result of the algorithm itself or the presence of a significant number of propaganda channels on the platform, whose activities are not moderated or labeled in any way.

  • This raises doubts, especially considering the platform’s subordination to a Chinese entity (ByteDance) operating within the Chinese legal system. Therefore, the app has significant potential to become a tool for shaping audience attitudes and views in the event of large scale influence operations.

  • Given the scale of Sino-Russian cooperation in information exchange, the TikTok app may pose a potential threat to the security of Polish users. Therefore, it is recommended not only to approach content appearing on the platform critically, but also to refrain from publishing sensitive data on the platform – this is particularly relevant for government entities and individuals associated with the state apparatus.

[...]

The analysis showed that the way content was presented on TikTok differed significantly depending on the topic. The highest percentage of pro-Chinese content was recorded for Xinjiang (58.7%) and Tibet (44.5%). This also applied to the issue of human rights in China (35%).

In turn, the results that referred to the Tiananmen Square massacre were dominated by content critical of the Chinese authorities (76.6%). "We treat such actions as a so-called "safety valve" – a strategy to create the impression of freedom of speech," the analysts write.

In the case of Taiwan, neutral materials predominated (92%). The most unusual results were obtained for the topic of re-education camps – as much as 89% of the displayed materials were not related to the searched topic. "The conclusions of the study clearly show that TikTok uses different content selection strategies for different topics.

"However, we should not focus solely on the operation of one algorithm. This is a question about who shapes the image of the world that millions of users see every day, and how they do it," says Antoni Mazek, an analyst at the NASK Disinformation Analysis Centre and co-author of the report.

[...]

In the flood of information, it is easier to give up further searches In the case of the topic concerning re-education camps, OAD analysts observed a large number of random videos. Unrelated content accounted for as much as 89%. In response to this specific query, several dozen materials appeared praising miraculous methods for weight loss (34% of content on the topic of health/weight loss), and even recordings of Chinese women doing the hula hoop.

"This is intended to discourage users from further searches – explains Dr Agnieszka Lipińska, head of OAD. "It simply causes frustration in the recipient. The search for information on an important topic collides with content that is less intellectually engaging," she adds.

Western influencers reinforce manipulation The analysed content rarely took the form of official messages. It often took the form of travel advertisements or lifestyle content. There was also material featuring Western influencers who constructed their narratives according to a similar pattern: they presented themselves as people "discovering the truth" about China and criticised Western media coverage. It was their materials that were most often recommended to Polish users. Prolonged exposure to such content may affect the perception of China by the audience, according to the researchers.

[...]

1646
 
 

Art by ShepGoesBlep Source

1647
1648
 
 

I engineered an independent AI platform that completely rewrites your resume for any open job description in under a second dynamically bypassing automated ATS filters Are you tired of facing instant rejections

Stop applying the old way

Spend 30 seconds to optimize your text Get your compatibility score in under a second Beat the system and stand out instantly Repeat this for every single job application

VortixData AI officially understands this nightmare and bridges the gap

Our pipeline adapts your resume to missing keywords instantly It evaluates your profile dynamically with an ATS compatibility score No credit cards needed try 5 parses completely for free Supports 8 global languages English Arabic Spanish Turkish and more

Cut down application time from 30 minutes to just 30 seconds

LAUNCH GIVEAWAY We normally offer 5 free credits but to celebrate our official launch I am permanently upgrading the token balance with massive free credits for the FIRST 50 USERS completely for free

Create your free account right now Drop your registration email here in the comments or send me a direct DM so I can personally boost your dashboard with your VIP tokens Let us save time and scale together 🚀🎯

1649
 
 

Edit: you all are getting rusty. :( It's okay, I'm getting old, too.

1650
 
 
view more: ‹ prev next ›