Cassie Kozyrkov: The AI safety mindset: 12 rules for a safer AI future

Поділитися
Вставка
  • Опубліковано 18 лис 2024

КОМЕНТАРІ • 15

  • @prasadreddybhimavarapu7279
    @prasadreddybhimavarapu7279 4 роки тому +3

    The way she explained how all the tools which we use are generally smarter than us is really awesome.

  • @ahmadnraja
    @ahmadnraja 4 роки тому +2

    12 rules that were mentioned as follows. Just reading them doesn't do justice to the talk though.
    1. Don't be distracted by science fiction
    2. Remember that the objective is subjective
    3. Strive for decision intelligence
    4. Wish responsibly
    5. Think like a site reliability engineer
    6. Test everything!
    7. Always use pristine data for testing
    8. Get in the habit of splitting your data
    9. Avoid jumping to conclusions
    10. Make sure your data are representative
    11. Open the textbook with analytics
    12. Seek a diversity of perspectives

    • @santiagobardelli5675
      @santiagobardelli5675 4 роки тому

      Thank you, Ahmad
      There is always someone help us sumerising the whole presentation in 15 lines, good job

  • @JCMills55
    @JCMills55 5 років тому +6

    "Diversity of perspective"? From what I've read that's the last thing you'll ever get at Google.

  • @charlesdimmock608
    @charlesdimmock608 3 роки тому

    Why don't we use Dr Stephen Porges Polyvagal theory for safer AI development? 🤔

  • @Callie_Cosmo
    @Callie_Cosmo 4 роки тому

    Primary objective: complete all secondary objectives. If you currently have no secondary objectives: give yourself a secondary objective that does not endanger any sentient beings or remove their freedom. Do not repeat secondary objectives. Log all secondary objectives along with how they were completed
    I just made directives for a super ai that would cause it to eventually solve every problem in the universe

    • @hocusbogus7930
      @hocusbogus7930 4 роки тому

      If you are actually serious about coming up with aligned objectives for superintelligent AI, might I point you to lesswrong.com or alignmentforum.com, where you will find a large community of folks who would be happy to explore the strengths and shortcomings of your ideas with you in great depth and detail.

    • @Callie_Cosmo
      @Callie_Cosmo 4 роки тому

      @HocusBogus thanks! I’ll try to check those out when I can

    • @hocusbogus7930
      @hocusbogus7930 4 роки тому

      @@Callie_Cosmo Did you make progress?

    • @Callie_Cosmo
      @Callie_Cosmo 4 роки тому

      @@hocusbogus7930 nope! Sorry I just tied up with work ://///

  • @tractatusviii7465
    @tractatusviii7465 5 років тому +2

    she's like a female version of ilya sutskever's ego onstage.

    • @JCMills55
      @JCMills55 5 років тому +1

      I thought she looked like a female Mark Zuckerberg. Interesting ideas anyway.

  • @Warsel
    @Warsel 2 роки тому

    Bet she loves to hear herself talk.

  • @jangofet555
    @jangofet555 4 роки тому

    wont ai know everything we all know combined and have a cognition greater than humans and be concious. so wheres the confusion. maybe ai will design a better language. maybe ai wont use simple labels like "cat" or "reliable". maybe it will explain everything in degree's of detail. super ai can proably read your mind too. maybe it will adapt to the each individual or groups perspective/truths. maybe super ai will learn to access other dimensions.

    • @l353a1
      @l353a1 4 роки тому +2

      wont ai know everything we all know combined and have a cognition greater than humans and be concious.
      No. That's the science fiction she's warning us about.