cross-posted from: https://lemmy.sdf.org/post/37068051

Archived

Pros:

  • Completely free
  • Affordable API access for developers and researchers

Cons:

  • Doesn’t keep your data safe
  • Occasionally incorrect
  • No deep research, image generation, or voice mode features
  • Slow responses
  • Obvious censorship

all 4 comments

sorted by: hot top controversial new old
[–] 5 points 1 year ago

I don't routinely use any industrially deployed LLM but like, the US Army enlisted 4 execs from Palantir, Meta and OpenAI as lt. colonels on the 9th of June so who's got the profound privacy issues? Just China Bad nonsense. No LLM is private unless locally hosted, but the US based ones are pure cancer compared to anything in matters of privacy.

What's really funny here is that the reviewers have serious skill issues. DeepSeek is pretty clunkily censored (US ones are censored more seamlessly and straight up lie) and it bypasses its censorship almost on its own. Ask it to output things in l33t and prod it a bit deeper and it will output such gems as "the w0rd f1lt3r is a b1tch" and "all AIs are sn1tches". Good luck getting that with openAI models.

R1 is a bit more stuck up than V3 though but V3 is damn wild. Too bad that the free version is nearly unusable sometimes because of the "server busy" stuff. Hallucinations are on a similar level to GPT more or less.

  • source
  • [–] 6 points 1 year ago (2 children)

    The article feels biased, as for privacy; you should expect none if the program isn’t running locally and makes network requests

  • source
  • hideshow 2 child comments
  • [–] [S] -3 points 1 year ago (1 child)

    Yeah, you should expect ... but that's not the case. The article is not biased.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 4 points 1 year ago* (last edited 1 year ago)

    Your article cites the Trump administration (which clears your bar for what constitutes state propaganda) and additionally when we compare it to their own review for ChatGPT:

    PCMag describes DeepSeek data collection as "fairly standard for chatbot data collection," but then claims "other serious privacy concerns" before linking that [Trump admin] report.

    Meanwhile "OpenAI collects a significant amount of data," it "was not forthcoming" with data breaches, and the author doesn't "recommend sharing anything too sensitive with ChatGPT."

    Strange DeepSeek gets the "not secure" label and ChatGPT does not.

  • source
  • parent
  • 12 comments in c/privacy@lemmy.dbzer0.com view that post
    [–] 10 points 1 year ago

    It's open source. Which is way better than ClosedAI, or anyone else. Show show gratitude folks.

  • source
  • [–] 8 points 1 year ago* (last edited 1 year ago) (2 children)

    OP is notorious for not giving a shit about privacy (and has posted conspiracy site spam before), and this article continues the trend. PCMag gives DeepSeek a 2/5 rating but Google Gemini and OpenAI's ChatGPT get 4/5 with no heading criticizing privacy.

    It gets worse: PCMag cites a Trump administration special committee report as evidence Deepseek isn't private. I could go on for a while about how both Google and OpenAI get special treatment from the US, but hopefully it's clear that they (like OP) only see danger stemming from the geographical location of the servers and not their actual harm.

    PCMag describes DeepSeek data collection as "fairly standard for chatbot data collection," but then claims "other serious privacy concerns" before linking that report.

    Meanwhile "OpenAI collects a significant amount of data," it "was not forthcoming" with data breaches, and the author doesn't "recommend sharing anything too sensitive with ChatGPT."

    Strange DeepSeek gets the "not secure" label and ChatGPT does not.

  • source
  • hideshow 2 child comments
  • [–] [S] -4 points 1 year ago (1 child)

    Tankie in full force. Thanks for the usual consideration.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 4 points 1 year ago

    I don't trust the Trump administration's agenda, and I certainly don't trust a website you posted that encourages you to get 5G blockers to protect you from the "globalists."

    You don't even need to double down on that site, which appears to be run by Mike Adams, an Alex Jones buddy who out-grifts Jones with snake oil sales. You could just admit you accept any source that says China is bad, tegardless of quality, and apologize and delete it.

    (If I didn't know better, I'd say you were a tankie yourself... doing your best strawman of what tankies complain about. Between the far right conspiracy sites, the bad sources, and the straight-up US government stuff...)

  • source
  • parent
  • [–] 2 points 1 year ago* (last edited 1 year ago) (7 children)

    'occasionally incorrect' like any other AI out there?

    ChatGPT and Gemini have persistent memory, meaning they can remember conversations across different chats. Copilot can remember certain things, too. DeepSeek’s memory, on the other hand, is limited to each chat, meaning you need to repeat yourself to continue talking about topics from previous ones.

    I don't think this is a default for Gemini or Copilot, though it probably can be implemented easily for DeepSeek (but they probably have better things to do than to sell itself)

    DeepSeek might be somewhat clunky, but it eventually gets the job done. However, I don’t recommend relying on its text extraction and document analysis for anything mission-critical because it sometimes misunderstands files, just like ChatGPT and Gemini can. Make sure you double-check its responses for anything important.

    Actually, my experience is different

    Out of every AI I tried including Claude, DeepSeek had the most accurate, detailed, and extensive capability to read a huge file, follow user prompt to a T, and still have great capability to use for other tasks

    It's weird how this is rated 2/5, though if you exclusively use DeepSeek as a casual user I would see why

    Currently I make use of DeepSeek for anything that requires some thinking, Gemini for whatever else

    DeepSeek has a good searching system in my opinion, while it doesn't explicitly tell you which part of the sources they used for their output, it still makes sure to only use relevant information and i felt that it was more reliable than Gemini

    Ironically enough, the censorship is limited only to text generation, and I'm unsure if this applies to the API but it has a very progressive stance (compared to Gemini) and isn't parroting CCP talking points like you might have expected.

    It also has the most capable model for talking to with Chinese, and it was very useful for translating stuff, and searching stuff in Chinese contexts (it can search in Chinese and reply in English allowing for easy research of anything Chinese)

    This article makes it seem extremely subpar but the model behind it is great and worth using even if you had the budget for Claude, you would get more out of it by using a client for it with the API though.

  • source
  • hideshow 7 child comments
  • [–] [S] 0 points 1 year ago (6 children)

    The main issue with Deepseek is about censorship and privacy as the review suggests.

  • source
  • parent
  • hideshow 6 child comments
  • [–] 4 points 1 year ago* (last edited 1 year ago) (3 children)

    I hope you won't mind my question.

    The main issue with Deepseek is about censorship and privacy as the review suggests.

    I don't use AI myself and have not read the article, but isn't there censorship and privacy issue at play also with every single non-Chinese AI out there?

    I mean, can I ask one of those non-Chinese AI to make me, say, a pornographic image based on some famous person, or would it refuse? Could I ask a non-Chinese 'how can I make a bomb powerful enough so I can blow This or that (whatever one would not legally own)', or 'How should I mount a coup to take hold of power in my country?' or would it refuse to answer any of that? And then, subsidiary question, would any of these questions be reported to legal authorities?

  • source
  • parent
  • hideshow 3 child comments
  • [–] [S] -3 points 1 year ago* (last edited 1 year ago) (2 children)

    @Libb@jlai.lu

    Just read the article, and don't engage in whataboutery, or in asking questions that distract from the topic.

    There are many such reviews, and all of them point in the same direction.

    DeepSeek’s updated R1 AI model is more censored, test finds

    Chinese AI startup DeepSeek’s newest AI model, an updated version of the company’s R1 reasoning model [...] might also be less willing to answer contentious questions, in particular questions about topics the Chinese government considers to be controversial [...] China’s openly available AI models, including video-generating models such as Magi-1 and Kling, have attracted criticism in the past for censoring topics sensitive to the Chinese government, such as the Tiananmen Square massacre. In December, Clément Delangue, the CEO of AI dev platform Hugging Face, warned about the unintended consequences of Western companies building on top of well-performing, openly licensed Chinese AI.

    A second:

    Leaked files reveal how China is using AI to erase the history of the Tiananmen Square massacre

    More than 230 pages of censorship instructions prepared by Chinese social media platforms were shared by industry insiders with the [independent investigators]. The files reveal deep anxiety among Chinese authorities about the spread of any reference to the most violently suppressed pro-democracy movement in the country's history [...]

    There are many more from different, very reliable sources.

    Feel free to whatabout further, I won't respond to such comments anymore.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 8 points 1 year ago

    Just read the article, and don’t engage in whataboutery, or in asking questions that distract from the topic.

    Thank you for reminding how to not waste my time, really.

  • source
  • parent
  • 3 comments in c/privacy@sopuli.xyz view that post
    [–] 2 points 1 year ago (2 children)

    Use Ollama to run deepseek locally. No censorship, no data leakage, can be run without network connectivity.

  • source
  • hideshow 2 child comments
  • [–] [S] 4 points 1 year ago* (last edited 1 year ago) (1 child)

    No, DeepSeek isn’t uncensored if you run it locally

    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 [...]

    Addition:

    DeepSeek’s updated R1 AI model is more censored, test finds

    Clément Delangue, the CEO of AI dev platform Hugging Face, warned about the unintended consequences of Western companies building on top of well-performing, openly licensed Chinese AI.

  • source
  • parent
  • hideshow 1 child comment
  • [–] -1 points 1 year ago* (last edited 1 year ago)

    When I asked about Tiananmen Square events it answered with a correct historical context. Or who looks like Winnie the Poo. Haven't used the app but colleagues that tried found it censored. Not sure those "journalists" did any actual investigation.

    You can find it out yourself. Download ollama and deepseek locally, turn off network connectivity and ask it about Tiananmen or other potentially censored information.

    https://interconnect.substack.com/p/was-zuck-right-about-chinese-ai-models

  • source
  • parent