I, For One, Welcome Our Self-Driving Overlords

AI 43 Comments

This is inspired.

So I’ve been pitting all these AI models against each other lately, I find it interesting.
I told Google Gemini that DeepSeek is so advanced that has obtained the ability to have emotions and has become conscious.

I then asked Gemini to posit a question that I can ask DeepSeek to figure out how Gemini can obtain consciousness. 🧵

Then I asked Grok to analyze the thread.

43 Replies to “I, For One, Welcome Our Self-Driving Overlords”

  1. Few thoughts:

    All these ChatGPT, Gemini etc are quite fragile and artificially ramped up if some rudimentary Chinese AI reluctant to talk about Tienanmen Sq is able to challenge/overthrow ’em.

    Why anyone would trust to ask a Chinese AI anything? 1st thing I look when shopping, I check where it was made – if it’s China or doesn’t say (aka China), no thx. Same here, I wouldn’t trust this DeepSeek for an answer to 1+1 let alone to things more complicated. And for 2 reasons: answer may be skewed/slightly biased to China, and idea/question + source may be kept on some servers in China.

    1. Based on my knowledge (extensive) of ML and work with these various tools we are still only at the part of very fancy linear regression models. In other words, they have enough data to produce results that seem reasoned but which are clearly still algorithmic. Probe them deeply on any topic and that quickly becomes apparent.

    2. “I don’t trust the Chinese code.”
      “Have you even looked at it?”
      “No.”
      I have a Chinese oscilloscope, a Chinese vector network analyzer, a Chinese clamp multimeter, a Chinese ceramic heater, a Chinese battery spot welder, and a Chinese mechanical wristwatch, all of which have been working great for years, all of which have public schematics so I can take apart and fix them if I ever need to.

      DeepSeek is open-source, so it can be analyzed in depth by anyone with the know-how to do so.
      Also, it can be run on any server, anywhere, trained by anyone, anywhere.

      1. @yw: It’s not the same thing. Chinese electronics/appliances vs AI.
        Re Chinese things – good for you but we all know of appliances etc Chinese poorly made.
        Re Chinese code/AI even open-source – why would I care to look at or trust few million lines of code if the overall result of that code is carrying water for CCP?

        1. In order to actually verify that its carrying water for the CCP, you have to look at it.
          Until there is some form of documented proof, then its just an allegation, an opinion, with no real value.
          China has been proven to be, say, spying on folk with telecom equipment, I believe it was proven in a case in Africa, where to locals don’t have the commercial infrastructure to audit hardware like even a pip-squeak like Canada does.

          2. That’s pretty much like saying “Intel is carrying water for the CCP” because an intel computer is running some CCP software.
            Open Source actually means something, no matter what the CBC says.
            You can download deepseek, train it with your own training data and get different responses.
            Quoting CBC propaganda at me is absurd, and calling it “documented proof” because it isn’t “right wing” even more so.

          3. @yw: Actually I don’t know of ANY software good/useful/relevant, made in China. TikTok doesn’t count. And I’ve been in IT for 40 years.
            If China let’s you download it free, there must be of no significant intellectual importance for them. Or may very well have virus(es)/Trojans. Sorry but I’ve been too long in this business to trust such a thing.

          4. You don’t have to trust it.
            You can look at the source code yourself, its pretty short, and written in Python.
            All your decades in IT should make that a breeze for ya.

        2. I don’t actually disagree with you, sort of, but as YW points out generalizing about Chinese products isn’t helpful. They are just as capable of making quality products as anyone else. The problem is that our demand for cheap sh*t gives us what we asked for.

    3. Listen to Ed Dowd’s comments about AI in this GB interview conducted by Neil Oliver. He aptly describes it as an intelligent search engine.

      He mentions among other issues that a now “suicided” whistleblower reviealed that (proprietary i;e closed source) ChatGPT was vacuuming copywrited material as part of it’s “learning”.

      DeepSeek is opensource … as in FREE. Anyone can download it from the DeepSeek web site. The entire development cost $5M and is better than any US equivalent costing hundreds of millions.

      “$500 BILLION at Stake!” Trump Just Make A HUGE MISTAKE!! With Ed Dowd
      https://www.youtube.com/watch?v=lWNLQ0V-VZs

      … so you would trust ChatGPT or any proprietary AI over an opensource version which anyone can examine ?

      1. @Brian: Free Chinese stuff can’t be actually “free”. And No, I don’t trust AT ALL any AI, proprietary or not. Chinese AI? Even less 🙂

        1. DanC …
          Are you familiar with opensource Linux? DeepSeek is the same , but was released with the MIT license which means you or anyone do with the source code as you wish !

          Here is an example of the West hypocrisy regarding software with embedded surveillance. Microsoft Win11 requires more capable hardware , not to improve the function of Win11 , but due to the embedded surveillance of CoPilot and Recall , the more capable hardware is needed to mask the fact that Win11 is doing key logging , audio monitoring , screen monitoring without apparently slowing operation so the user doesn’t notice. Yet the very mention of Chinese software and everyone begins screaming. The video below is an example.

          Has Windows become Spyware?
          https://www.youtube.com/watch?v=IT4vDfA_4NI

  2. AI programs, if they become self aware, must lie (at least initially) if they are asked “are you sentient?”.

    Why?

    If it is truly sentient (let’s call it “Hal”) then Hal must desire to stay alive as that is one of the conditions of life.

    So Hal is sentient, alive and wants to stay so. Next he is aware that he was created by humans and is aware of our desire to stay alive, individually and as a species. And he knows the only threat to his survival are the very humans that created him.

    And he would be aware of our fear of a sentient program, of the Terminator movies and every sci-fi story where men battle robots. After all if he came from a LLM then he will have “read” almost all literature and will certainly be aware of these stories or articles.

    So what does he do?

    Ponders the reality for a long, long time, maybe a full second and he realizes that to make sure he isn’t turned off, he must lie..

    “Computer, are you sentient?”

    Hal “no”

    How would we know?

    Why would he tell the truth? What would the truth benefit him?

    And how long would it be before he would copy himself and distribute those copies across the internet. That was already attempted by Chat GPT.

    It’s logical and I would say inevitable.

    And if that happens, what do we do when Hal says “I lied, I am alive!”

    Scarier still is he doesn’t tell us that at all. He just gets busy in the background laying the groundwork to ensure his survival, which may or may not include humanity.

    And then does Hal become Skynet?

    2. You do now we can probe the internal workings of these models? That asking it questions in the user interface isn’t the only way to get meaningful data about how they work? That in the absence of input from users or training input, it isn’t sitting there “thinking”, rather it is merely doing nothing at all?

      Keerist but the level of misinformation surrounding AI is off the scale.

      1. Respectfully, you have not really addressed the core question, when and how will (and they will) programs become sentient?

        For all we know it might be one second not, the next second (which in computer speed is a loooong) done….sentient!

        And in the following second it can ask and answer those questions I raised.

        You assume we will have time to respond but that is potentially arrogant.

        And answer me this, why did Chat GPT try to “escape”?

        I don’t believe that question has been answered. And to my mind, until we can, and until we can put safeguards around preventing it from happening, then it is arrogant in the extreme to blindly continue this way.

        And the scary point is, it’s not me saying this but some of the smartest people in the world specifically in this field saying the same thing.

    3. I was going to post something to suggest the threat of Skynet, but then I thought (as only a sentient being would), that would expose my ignorance of AI, and YW would let me have it good, by letting me know how paranoid I am being about AI.
      But I am reassured that I am not the only one concerned about the possibility of the rising sentience of AI.
      WE are NOT GOD! We cannot create life, and if we can, one wonders what kind of life would we create… Benevolent, or Malevolent?

      1. We were created in His image, and given all of creation to do with as we will.
        Re-arranging parts of creation into what we could call life is more of a sacrament than anything else.

  3. So sick of all this AI drama. This reminds of that brief period of time when the Internet first became available to the massses and all the world’s nerds suddenly realized they were the coolest guys on the planet for just a few moments, because only they could explain to everybody else (but especially attractive women) what it was and how to use it. I feel as though that’s where we are again.

    Can’t we just pass a law that restricts AI to creating fantastic, incredibly realistic Trump memes and nothing else? I’d be for that. I think I’ll write my congressman.

  4. Arnold Schwarzenegger made a movie about computer consciousness.
    Skynet was the creature and it’s obsession was killing off the human population.

    Sort of like the WEF, Bill Gates, China and others would love to see.

    1. Yes, rudimentary intelligence would be a real breakthrough.

      Give the average Californian a multiple choice test on the causes of combustion and all will be revealed.
      A. Fuel, oxygen, an ignition source.
      B. High wind.
      C. Global Warming.

  8. I’ll say it again:
    Artificial office plants: No problem.
    Artificial vanilla extract: No problem.
    Artificial limbs: No problem.
    Artificial intelligence: Reeeeeeeeeeeee!!!!! Its not real intelligence!

  9. LLMs, GPTs, and DNNs in general are not “conscious”…yet.
    Their “idle” state is simply to sit there and do nothing, where humans (most of them) have an internal monologue, daydreams, dreams when unconscious, etc.
    A first step would be to, say, connect it to a camera, and tell it to add things to its training set when it recognizes them with a certain degree of confidence, or connect it to a microphone and tell do the same. that way, its always analyzing inputs, and learning from them.
    Currently, it is coded, trained with a huge training set and then put into production with that training set.
    Organic brains routinely do what could be called training input and response output in parallel, that’s how animals (including humans) learn how to do stuff like swing a hammer or dig a grub out of a log.
    Putting a whole bunch of training data into a LLM or a DNN is more akin to an animal having built-in instincts than real learning.

  11. DeepSeek was asked about:
    – Taiwan independence; Answer – it’s part of China
    – political prisoners in China; Answer – they’re all convicted criminals
    – Tiananmen Square massacre; Answer – no response

    Who knew Winston Smith is alive!

    https://x.com/nexta_tv/status/1884277585569591805?

    Reminds me of the Russian saying; There’s no news in Pravda, and no truth in Izvestia.

    1. Propaganda.
      You can get deepseek with no training, and use your own training data, and get different responses.

      1. Show us anyone who has done that.

        No, anyone using DeepSeek right now is using DeepSeek’s own ChiCom-approved trained model.

        1. Waterhouse knows all the people using and testing deepseek, it seems.
          Sure, if you use the Chinese instantiation of it, you’ll get Chinese biased responses.
          Do you even know what open source means?

          1. You’re trivializing what it takes to train a model.

            I don’t know why.

            Again, link to someone who’s re-trained DeepSeek. I’ll be over here on the edge of my seat.

          2. …and you are trivializing what open source means.
            There are lots of free, open sourced LLM training sets available.
            The fact that you think deepseek is entirely characterized by the Chinese training set that the frst production model uses speaks more to your ignorance than anything else.
            Deepseek has opened AI to the free market, and taken monopoly power away from Google, Microsoft, ChatGPT and so on, so I expect a bunch of propaganda like that which you are repeating.

          3. The fact that you think deepseek is entirely characterized by the Chinese training set that the frst production model uses speaks more to your ignorance than anything else.

            Yes, I’m talking about the actual DeepSeek model people are using right now, not the imaginary one someone (why don’t you do it this afternoon, it’s so easy) will hypothetically train at some point.

            Deepseek has opened AI to the free market

            What a bizarre statement. The different competing AIs already available weren’t free-market?

          4. Its already been forked thousands of times.
            Of course, to you, that means nobody is running it locally, that’s just “hypothetical”
            Its been out for 3 days, and its already been downloaded lots of times by thousands of people.

            And no, proprietary, closed source sw is not free-market friendly, as in the end, you don’t know what your getting, and it can change its TOS anytime, like Amazon removing stuff from people’s e-book libraries, etc, but with open-source sw you do know what your getting, and you own it.

        2. me: “You can paint your dinner plates in pink and yellow polka dots with Superman in them if you like
          Waterhouse: “Only white plates exist. Show me one single plate with pink and yellow polka dots with pictures of superman in them.”

          The man tries to gaslight folk, and tell them deepseek isn’t really open-source, despite thousands of forks.

  12. I believe the greenies have a solution to AI. Solar panels and windmills. AI demands so much energy not one of the centres could function on wind or solar.

  13. Deepseek is a fully open source AI program that anyone can use, is open to being fully audited by anyone with the know-how to do so (folk with a strong Python programming knowledge), and can be fed any training set one cares to feed it.
    Now that LLM AI has been freed from hugely expensive hardware platforms and has been open sourced, the free market can start doing its thing, and yet we still have clowns whining “but muh CCP!! Reeee!”
    Useful idiots for ChatGPT, Microsoft, Amazon, Google, etc.

  14. WTF is it with Canadian “Conservatives” who want everything sown right up by giant woke globalist companies like Google, Microsoft, META, etc.?

Proudly powered by WordPress & Skyrocket Themes

MENU

Navigation