Victoria Krakovna

@vkrakovna

Research scientist in AI safety at DeepMind. Co-founder of Future of Life Institute . Views are my own and do not represent DeepMind or FLI.

Vrijeme pridruživanja: lipanj 2014.

Tweetovi

Blokirali ste korisnika/cu @vkrakovna

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @vkrakovna

  1. proslijedio/la je Tweet
    23. sij

    Thanks to structural causal models, we now a more precise understanding of incentives in causal influence diagrams blog post: arXiv:

    Poništi
  2. proslijedio/la je Tweet
    5. sij

    RL shifts the question of what intelligent behavior is to finding a reward function. I think we should focus more on what environment and reward function rather than on what RL algorithm to use. Is there theory for how properties of env and reward affect the resulting behavior?

    Poništi
  3. proslijedio/la je Tweet

    Join us for our final podcast of 2019 with and on consciousness, ethics, effective altruism, human extinction, emerging technologies, and the role of myths and stories in fostering societal collaboration and meaning.

    Poništi
  4. 20. pro 2019.

    My current thoughts on the specification gaming examples list (1.5 years after its release)

    Poništi
  5. proslijedio/la je Tweet

    In this episode of AIAP, Jan Leike discusses his movement from theoretical to empirical AI safety research — why empirical safety research is important, how this has lead him to his work on recursive reward modeling, and the work being done at DeepMind.

    Poništi
  6. proslijedio/la je Tweet
    13. pro 2019.

    We present ReQueST: a method for training RL agents from human feedback in the presence of unknown unsafe states. By , , , , Paper: Code:

    Prikaži ovu nit
    Poništi
  7. proslijedio/la je Tweet
    13. pro 2019.

    How do you train an RL agent in the presence of unknown, unsafe states without visiting them even once? New algorithm by our intern synthesizes trajectories with a generative model and ask a human to label them for safety.

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet

    Stuart Russell discusses his newest book on the AI Alignment Podcast, Human Compatible: Artificial Intelligence and the Problem of Control.

    Poništi
  9. 4. lis 2019.

    Open-source code for the paper "Penalizing side effects using stepwise relative reachability", comparing different design choices for side effects penalties:

    Poništi
  10. 21. kol 2019.

    The fourth episode of the new podcast discusses specification problems in AI safety, Goodhart's Law, and reward learning.

    Poništi
  11. 21. kol 2019.

    New blog post on classifying AI safety problems as different types of Goodhart's law

    Poništi
  12. 16. kol 2019.

    If you're interested in 's safety work, it is now all in one place on the new website

    Poništi
  13. 14. kol 2019.

    Exciting work on the reward tampering problem in AI safety, where the agent changes its reward function by exploiting how reward is implemented in the environment. The paper proposes design principles for building agents without an incentive to tamper with the reward function.

    Poništi
  14. proslijedio/la je Tweet
    11. srp 2019.
    Poništi
  15. 21. lip 2019.

    New paper on modeling AI safety approaches with causal influence diagrams

    Poništi
  16. proslijedio/la je Tweet

    FLI's co-organized the Safe Machine Learning workshop at . Read her recap of the event for an overview of the talks, panels, and papers.

    Poništi
  17. 17. lip 2019.

    My belated summary of the Safe ML workshop at ICLR

    Poništi
  18. proslijedio/la je Tweet
    9. lip 2019.

    Very excited to deliver the tutorial on tomorrow together with ! Be prepared for fairness, human-in-the-loop RL, and a general overview of the field. And lots of memes!

    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    3. lip 2019.

    DeepMind’s spoke to about the importance of building robust and safe AI systems. Listen to the full podcast to find out why a career in machine learning makes a difference:

    Poništi
  20. 19. svi 2019.

    Separate recordings of SafeML talks are now available on the workshop website

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·