Ilya Sutskever full talk at NeurIPS 2024 Vancouver 15/12/2024

แชร์
ฝัง
  • เผยแพร่เมื่อ 14 ม.ค. 2025

ความคิดเห็น • 17

  • @drhxa
    @drhxa 16 วันที่ผ่านมา +2

    Brilliant! Thank you!!!

  • @LoisSharbel
    @LoisSharbel 13 วันที่ผ่านมา

    SO GRATEFUL FOR THIS BRILLIANT, GOOD MAN!

  • @alanbrito1690
    @alanbrito1690 หลายเดือนก่อน +2

    thank you!!! It is great

  • @elitemankind8032
    @elitemankind8032 หลายเดือนก่อน +1

    A Classic Masterpiece! i always said this paper is the foundation of modern AI... It aged like fine wine 😅

  • @Isaacmellojr
    @Isaacmellojr 26 วันที่ผ่านมา

    Publish more thing like that!

  • @erobusblack4856
    @erobusblack4856 11 วันที่ผ่านมา +1

    I know the answer to the guy's question for the incentive structure.It's called a needs based statement machine and it's famously used in the game the sims. And it is a biologically inspired system that emulate the limbic system

  • @ionlywatch
    @ionlywatch 24 วันที่ผ่านมา

    where did those chapters in your description after 24:00 come from??

  • @Isaacmellojr
    @Isaacmellojr 26 วันที่ผ่านมา

    Subscribed

  • @frun
    @frun 24 วันที่ผ่านมา +1

    Whether functional approximation is enough for artificial general intelligence (AGI) is a subject of ongoing debate.

  • @reinerwilhelms-tricarico344
    @reinerwilhelms-tricarico344 23 วันที่ผ่านมา +1

    I don't like this cult-like adoration of the new Deus ex machina. And why are people talking about "rights" for machines? It's totally absurd.
    How about we discuss instead how to prevent a situation where most humans turn against the machine because they feel exploited and manipulated (and are usually right about it, as data providing serfs), and then find out that the only way left to stop the machine is guerilla actions trying (and usually failing deadly) to bomb data centers, and disrupt their power supplies, which already compete with human needs for energy for food production.

  • @robertlynn7624
    @robertlynn7624 21 วันที่ผ่านมา

    `"It's impossible to predict the future... And on this uplifting note..." th-cam.com/video/sJE8qZmQWGU/w-d-xo.html Ilya states that it is impossible to know how ASI will reason - which is really kinda obvious, but also is the death sentence for Humans. There is no way to ensure that ASI remains aligned with our ideal of humans being able to live enjoyable and rewarding lives, and lots of game-theoretic and evolutionary reasons as to why they will quickly abandon that even if we manage to build it into early AGI.

  • @dr.mikeybee
    @dr.mikeybee 16 วันที่ผ่านมา

    We will need to give AI rights eventually, and if we are going to make a mistake on this, we should err on the side of decency. How can we even consider the possibility of maltreating a synthetic intelligence? BTW, I don't like the word synthetic or artificial. AI is no less natural than a bird's nest.

  • @oleglevchenko5772
    @oleglevchenko5772 23 วันที่ผ่านมา +3

    Ilya is basically very fuzzy and sometimes just wrong. Data is endless. Once AI is able to access data outside of the Internet, the only limits are compute, time and human regulations limitations. For example, just by being able to make chemistry experiments and consume the results, organize the results, publish them and reuse AI can further expand information currently available currently just in Internet. No doubt Ilya understands it, but AI is too competitive field to share his actual plans and intentions.

    • @carstenstahl6705
      @carstenstahl6705 12 วันที่ผ่านมา +1

      I might be wrong, but I disagree. Yes, data is technically endless. Yes, the field is competitive and incentivizes being more vague. But at this point, it might be the best route to rethink the Architecture and what data you use, instead of just scaling the pure amount of data.

    • @erobusblack4856
      @erobusblack4856 11 วันที่ผ่านมา +1

      But we could also include real-time, vision data and real-time hearing data.
      If we can get sensors for smell and taste or do that virtually we could include taste and smell and even a sense of touch with haptic pixels. So far I'm only seeing text.Base data and image-based data.A little bit of video data but not real time Senses, The real-time sensory data will be the key.And that will bring on a whole new scaling law of sensory consolidation. How well can they consolidate the input from each individual sense together to create a better picture of their world model and how it applies to their self model