Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
AI

Taiwan Says Government Departments Should Not Use DeepSeek, Citing Security Concerns (reuters.com) 37

An anonymous reader shares a report: Taiwan's digital ministry said on Friday that government departments should not use Chinese startup DeepSeek's artificial intelligence (AI) service, saying that as the product is from China it represents a security concern.

Democratically-governed Taiwan has long been wary of Chinese tech given Beijing's sovereignty claims over the island and its military and political threats against the government in Taipei. In a statement, Taiwan's Ministry of Digital Affairs said that government departments are not allowed to use DeepSeek's AI service to "prevent information security risks".

"DeepSeek's AI service is a Chinese product, and its operation involves cross-border transmission and information leakage and other information security concerns, and is a product that jeopardises the country's information security," the ministry said.

Taiwan Says Government Departments Should Not Use DeepSeek, Citing Security Concerns

Comments Filter:
  • by Rei ( 128717 ) on Friday January 31, 2025 @11:08AM (#65132625) Homepage

    You can download .safetensors versions and run it yourself; safetensors files are just numbers, not code. And if the fear is of the fact that the Great Firewall is baked in, you can ablate it (the Qwen 32B distilled version of DeepSeek R1 has already been ablated).

    • You can download .safetensors versions and run it yourself; safetensors files are just numbers, not code. And if the fear is of the fact that the Great Firewall is baked in, you can ablate it (the Qwen 32B distilled version of DeepSeek R1 has already been ablated).

      What else besides safetensors is needed to get the same ChatGPT-like user-level functionality? It would be interesting to see how much of the Chinese-mandated censorship comes across with safetensors.

      If at least part of the censorship is bypassable using safetensors or by otherwise running completely locally, it would be interesting to see the Chinese government reaction to social media posts showing DeepSeek models flouting Chinese censorship.

      • Since a substantial percentage of the tokens are from Chinese media and sources, it will have the same biases seen in those sources. Just as sources trained on US media will have a US bias.

        Of course to the extent that Chinese media publish propaganda the model will learn that propaganda (similar to US media publishing of propaganda), sometimes the US reporting will eventually correct the propaganda though most media don't bother.

        As to deliberate censorship - the model itself appears to not be censored but r

        • by Rei ( 128717 )

          Since a substantial percentage of the tokens are from Chinese media and sources, it will have the same biases seen in those sources.

          The base model is actually quite well aware of Chinese human rights abuses.

  • by Virtucon ( 127420 ) on Friday January 31, 2025 @11:08AM (#65132627)

    Ya think?

    But then again who can you trust?

    • by Tarlus ( 1000874 )

      I've been saying this since the advent of Siri et al. I wouldn't trust that with any information I wouldn't want the wide world to know, much less AI chat bots housed in the United States.

      • I have a friend that wired his house up with Alexa smart devices, light switches etc. It is always disconcerting when I go to his home and he or his wife/extended family will ask Alexa to turn the TV on or Alexa do this or that. It makes me honestly uncomfortable to say anything. Then again his lifesize Mitt Romney cardboard cutout is creepy AF too.

        • by mysidia ( 191772 )

          We need a new version of Alexa.. perhaps a custom user mod: Where the microphone is electrically disconnected by a physical switch, except when you are holding down a button on the device that indicates you want to talk to the agent.

          • Uh no. No one needs Alexa.

            And you think the privacy fix is to have to physically touch a device to then be able to tell it to turn on a light or the tv via voice?
            Wut? Why not just turn on the light yourself?

            No one who isn't disabled needs a voice activated light switch or anything else Alexa does.

      • by mysidia ( 191772 )

        My understanding's at least with Siri Apple would not permit even their internal admins and devs who were troubleshooting issues with that app the capability to find out the User queries.

        Whatever you told to Siri is probably safer than the contents of files stored on your cloud drive. Or on your local hard drive; once you realize that your OS runs an antivirus program, and those antivirus programs sometimes to decide random files should be submitted to a server to be checked.

  • by jfdavis668 ( 1414919 ) on Friday January 31, 2025 @11:24AM (#65132645)
    I think we should take its opinion into account.
  • by waspleg ( 316038 ) on Friday January 31, 2025 @11:26AM (#65132647) Journal

    getting downvoted for asking if anyone had actually run Wireshark to see if the "local" version was actually sending shit back or not. No one answered yes or no.

    I do not understand why anyone default trusts a single thing which comes out of China with obvious ties to the CCP.

    • by Virtucon ( 127420 ) on Friday January 31, 2025 @11:38AM (#65132669)

      I agree.

      Curiously as an analog, how many millions of people use TicTok? [statista.com]
      If you look at that data you'll see a curious omission, no data from China. So an app that funnels data to the Chinese government isn't being used in China or they're purposely omitting the data from public scrutiny.

      • by karmawarrior ( 311177 ) on Friday January 31, 2025 @12:43PM (#65132839) Journal

        Or TikTok isn't available in China, which makes sense because any social network in China is going to run into problems with censorship.

        And that's exactly what happened: TikTok isn't available in China. No conspiracy theory needed.

      • The link you provided is subscription-walled so I can't look at the data. But I can tell you that TikTok in China is VERY different from what it is in other parts of the world: https://www.deseret.com/2022/1... [deseret.com]

        Relevant quote from the above link:

        “It’s almost like they recognize that technology is influencing kids’ development, and they make their domestic version a spinach version of TikTok, while they ship the opium version to the rest of the world,” Tristan Harris, a former Googl

        • What your world will look like: lots of disappointed kids. Most of them will never be either. Guess which one is more attainable?

          • The alternative to social media influencer is only fans girl.
            The alternative to astronaut is scientist.

            Guess which one is more attainable?

            • With these educational costs? Get a clue, noob.

              • I take your point, but would add that high educational costs both cause, and are partly caused by, a dumbing down of the populace. Also, if social media is a type of propaganda - and I think it is - then I'd rather it be propaganda that reinforces aspiration and fosters self-worth. If we want young people's reach to exceed their grasp, I think ending up lower on the 'astronaut' ladder is preferable to anywhere on the 'influencer' ladder.

                • I take your point, but would add that high educational costs both cause, and are partly caused by, a dumbing down of the populace.

                  This is absolutely true, but it's also absolutely intentional, and we know whose intent it was. And until we get their hands off the yoke (and just look at how that's going right now) our steep dive will continue.

                  If we want young people's reach to exceed their grasp, I think ending up lower on the 'astronaut' ladder is preferable to anywhere on the 'influencer' ladder.

                  It definitely is, but we're allowing the future of those jobs to be destroyed right now, and it's not just in the USA. There's a rising tide of fascism across most of the developed world, and it doesn't tolerate knowledge or caring about facts.

    • Isn't it open source? Wouldn't it be relatively obvious to anyone examining the source code if there was a back door?

      Is it just me or does Slashdot's anti-China community seem a little OTT these days? I'm not a fan of the regime but it feels as if an article talking about new smart ceramic cups "made of china" would get flooded with comments concerned about whether the CCP is poisoning your tea or something...

    • I do not understand why anyone default trusts a single thing which comes out of China with obvious ties to the CCP.

      I do not understand why anyone default trusts a single thing which comes out of any major corporation or government. I doubt and question all of the software I use, because it all runs on devices which have or may have internet access.

    • by gr8dude ( 832945 )

      My understanding is that the model is not a program that can execute any logic on its own, open network connections, etc. The model is a huge data file, which is then loaded by another program (like Ollama or llama.cpp) that deals with the user.

      Some software that uses neural networks has functionality to perform HTTP requests and fetch some data that are then used to enrich the prompt for the model (e.g., OpenWebUI, if I recall correctly), or execute commands in the system (e.g., the Cline plugin for VSCode

  • by ctilsie242 ( 4841247 ) on Friday January 31, 2025 @11:27AM (#65132653)

    I wonder about just running it locally, where it doesn't send any training data to the mother ship. Downside is that it does require some beefy hardware to run (although it can give queries on a M1 Mac).

    • WTF you need random software downloaded from the internet in a government agency? I am tired of getting through courses telling me "push the blame to IT", "do not be liable for any hack".
      • by mysidia ( 191772 )

        The software serves a purpose and it may be extremely helpful.

        I think you just need to implement better segregation inside your endpoints - which the Qubes os does just fine. Basically segregated (virtual) systems that run your "random software" which have no access to your secure network, and no access to the data you are required to protect.

        For this case; each random program should have a segregated compute instance and a segregated network instance with no access to other applications' network. An

  • by hydrodog ( 1154181 ) on Friday January 31, 2025 @12:56PM (#65132875)
    I assumed that open source meant that the code for deep seek is open source, and that the procedure to train it is laid out: Here is the terabytes of data, and you run the training algorithm and out pops deep seek 7 If that's not true, and only the code is open, then that barely counts because a vast majority of the result is the training set. It makes sense that it's not open, because if it were open, there wouldn't be arguments about the training data scraped from chatgpt. They would simply have each question and corresponding answer, and that would be the training data. It wouldn't be in dispute.
  • by thtrgremlin ( 1158085 ) on Friday January 31, 2025 @01:35PM (#65132985) Journal
    Big Brother distrusts creepy uncle.
  • Beijing's sovereignty claims over the island

    But while we read that, we must remember that almost all nations in the world consider Taiwan to be part of China [wikipedia.org]. Even the USA does, at least for now.

Real Users know your home telephone number.

Working...