>AGI is 5-10 years away

>AGI is 5-10 years away
What is this prediction based off of? Is there any conceivable way forward for general reasoning at the moment? Are people just being fooled by LLMs?

Nothing Ever Happens Shirt $21.68

The Kind of Tired That Sleep Won’t Fix Shirt $21.68

Nothing Ever Happens Shirt $21.68

  1. 1 month ago
    Anonymous

    At this pace of 15% cpu improvements a year we'll only have 2x'd our hardware abilities.
    Might be possible with npu's they're incredibly efficient at the Ai tasks theyre made to execute

  2. 1 month ago
    Anonymous

    >line goes up
    >AGI soon brother
    Pure moronation.

  3. 1 month ago
    Anonymous

    troony picture

    • 1 month ago
      Anonymous

      homosexual post.

  4. 1 month ago
    Anonymous

    >Are people just being fooled by LLMs?
    If an LLM can fool you to think it's a form of general intelligence ... is it that really "fooling" anymore?
    You call it trickery only because you know how it works on a surface level.

    Either way we're at a point where mere performance improvements wont get us there. The recent AI boom came about from research that figured out how to trick these neural networks we already had to do our bidding.
    All AI progress is and will be will be architectural. Does an AGI-capable architecture exist, that can be trained with the datasets we have (the internet)? That is the gorillion dollar question.

    • 1 month ago
      Anonymous

      Well, the people actually developing the models have a limited understand of how they work, so really, nobody can predict how this technology will develop or how quickly. I mean, the architecture was discovered entirely by accident. I think we are in one of those woeful situations where the technology outpaces understanding of the technology.

    • 1 month ago
      Anonymous

      >If an LLM can fool you to think it's a form of general intelligence ... is it that really "fooling" anymore?
      Black person, go to GPT-4o and try to have a realistic human conversation for 5 minutes. When you hear what it outputs with the sound and cadence of human speech it makes it's restrictions ridiculously obvious. I promise it will dispel any mysticism you have over this crap.

  5. 1 month ago
    Anonymous

    Maybe combining multiple data input, like sounds, text, 3d space, temporal coherence, can make emerge some kind of complex understanding ?

    • 1 month ago
      Anonymous

      The current thinking is that in order to have an intelligence that in every way mimics what we'd consider "intelligent" it has to have a body via which it can interact with the world.

      • 1 month ago
        Anonymous

        ewww sentient drones

  6. 1 month ago
    Anonymous

    blind optimism. llama 3 got it's big improvement by scaling the amount of tokens trained on by 7 times to 15 trillion tokens. There is supposedly some type of multi modal architecture in the works like what gpt4o is doing but it is not the massive push people are looking for.

  7. 1 month ago
    Anonymous

    AI research in non linear. At least that;s the thinking of the guy from OpenAI on Dwarkesh Patel last week. GPT-4o is no longer an LLM so much as a hybrid.

    I still think Yann LeCun makes more sense, but there are uncertainties and unknownsahead and you'd be a fool to bet otherwise.

  8. 1 month ago
    Anonymous

    >bro trust me, just a million more parameters and this text processor will spawn consciousness i swear

  9. 1 month ago
    Anonymous

    based on marketing from tech companies and AI startups.

    the fact is we currently have a bunch of shitty bots that parrot wikipedia articles at you and are easily confused and led to contradict themselves. morons are calling it AGI because they heard the term somewhere and think it makes them sound smart and in the know. other morons are predicting the end of civilization because of chatbots that speak like a redditor and don't understand logic puzzles.

  10. 1 month ago
    Anonymous

    >>AGI is 5-10 years away
    Two more weeks!

    homie, I've been told in 2023 that there will be the end of the world by December, when ChatGPT 5 will be released. It didn't. It won't ever be, probably.

    The biggest bottleneck to "agi" is power consumption. Chatgpt eats about 1 GWh/day, which is daily consumption of roughly 33000 US households. AGI trannies mostly think about computational power, and that we need more data and shit, but in reality it's all about power consumption. And unless there's a major leap in that area, AGI dream is dead.

    ?si=8XedvhUp-r1eKm0F

    • 1 month ago
      Anonymous

      So that's why bill is trying so hard to shill nuclear energy.

  11. 1 month ago
    Anonymous

    >we'll suddenly start seeing magical leaps in solar technology
    >because ML is out-scaling current energy tech

  12. 1 month ago
    Anonymous

    bend over

  13. 1 month ago
    Anonymous

    >AGI finds the furry genome sequence
    >gradually kill off blacks and replace them with test tube anthro women

    paradise is coming.

Your email address will not be published. Required fields are marked *