Connor Leahy

@NPCollapse

Hacker - A(G)I, neuroscience, security, philosophy - Slow or fast takeoff, in 2045 or 2200, Ems or AIs, it doesn't matter. AI Alignment is a hard problem.

Munich, Germany
github.com/ConnorJL
Vrijeme pridruživanja: siječanj 2019.
5 fotografija ili videozapisa Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @NPCollapse

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

  1. Prikvačeni tweet
    20. stu 2019.

    3 korisnika označavaju da im se sviđa
    Poništi
  2. proslijedio/la je Tweet
    2. velj

    im going to shut down for a bit while you all figure out how to make your computers sexier. come back when the internet is sexy

    7 replies 276 proslijeđenih tweetova 1.402 korisnika označavaju da im se sviđa
    Poništi
  3. 31. sij

    I think this may be the best distillation of the effective altruist/rationalist/whatever mindset. I feel almost understood. It's so strange for me that other people aren't as fundamentally obsessed with morality/suffering as I am.

    2 korisnika označavaju da im se sviđa
    Poništi
  4. 31. sij

    A huge thanks to and the other people that have helped to improve these essays. Couldn't have done it without you!

    1 reply 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  5. 31. sij

    I know my essays kinda went a bit off the rail from the start, but I'm happy to finally have pushed through to the end. In the future I may return to them and refactor them properly. For now, I'm excited to move on to new blogging adventures, so stay tuned!

    1 reply 1 korisnik označava da mu se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  6. 31. sij

    "I don't care if he was made of metal, he was my friend." I have published the fourth and final part of my "Counting Consciousness" series. This time, I go full philosophical and finally explain what "counting consciousness" means!

    2 proslijeđena tweeta 5 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  7. proslijedio/la je Tweet
    29. sij

    um.. google's latest chatbot is 😳

    297 proslijeđenih tweetova 1.238 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  8. 29. sij

    btw I learned of this and other cool advancements in cryptography (directly or indirectly) from here:

    1 korisnik označava da mu se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  9. 29. sij

    Zero Knowledge Proofs allow you to prove you know something to someone else without sharing the secret knowledge you have. It's not only totally non-obvious to me that that is even possible, but in fact ANY proof can be zero knowledge! Mind blown!

    9 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  10. 29. sij

    Oh boy

    New paper: Towards a Human-like Open-Domain Chatbot. Key takeaways: 1. "Perplexity is all a chatbot needs" ;) 2. We're getting closer to a high-quality chatbot that can chat about anything Paper: Blog:
    Prikaži ovu nit
    1 reply 2 proslijeđena tweeta 4 korisnika označavaju da im se sviđa
    Poništi
  11. 22. sij

    b) If we don't: How do we measure utility, how do we deal with pascalian mugging, how do we count/weigh nonexistant "people", etc etc. Basically all problems of utilitarianism, and there's a lot of them. The question quickly becomes incoherent if you don't make strong assumptions

    1 reply
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  12. 22. sij

    ...is actually at some "repugnant" level. It might occur at some comfortable midpoint between population size and wealth. This seems to be the case at least intuitively.

    1 reply
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  13. 22. sij

    For those interested: My reply to TRP is a) If we just accept all TRP assumptions (especially that utility can be measured cardinally and added): It's not a priori clear that the "set point" where adding one more person will cause more harm than good...

    1 reply
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  14. 22. sij

    Speaking of Parfit, here's a pretty great couterargument to the Repugnant Conclusion. It isn't the first one I would have thought of, because it relies on some fishy metaphysics, but so does TRP.

    1 reply 1 korisnik označava da mu se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  15. 22. sij

    Parfit is one of the philosophers I am most interested in. I've always been weirdly obsessed with moral philosophy, and he seems to take it similarly seriously. I've had all his books on my shelf for weeks now, but haven't been able to tackle them yet

    1 reply 2 korisnika označavaju da im se sviđa
    Poništi
  16. 21. sij

    The most surprising part of the audiobook though was probably that it had its own rap song. (???) Does Cambridge Audio frequently commission dank science rap songs?

    1 proslijeđeni tweet 3 korisnika označavaju da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  17. 21. sij

    I've read "The Ape That Understood the Universe". It was a good book with a fun writing style that in many ways is "The Selfish Gene 2.0: Human Sexuality Boogaloo". If that's your thing, you'll probably like it. I also liked the Appendixes on how to win arguments with deniers.

    1 reply 2 proslijeđena tweeta 21 korisnik označava da mu se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  18. 18. sij

    It begins. Fake Twitter propaganda accounts using AI generated faces.

    Per , there's always (at least for now) a glitch in the AI-generated avi matrix
    Prikaži ovu nit
    1 reply 2 proslijeđena tweeta 4 korisnika označavaju da im se sviđa
    Poništi
  19. proslijedio/la je Tweet
    14. sij

    Philosophers of mind like to present themselves as exploring the relationships between mind and world. What they actually do is more like trying to assemble a broken and incomplete Ikea set, with instructions passed down by an elaborate centuries-old game of telephone.

    12 proslijeđenih tweetova 100 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  20. 15. sij

    Deep Double Descent: Training a SGD model after it hits 0 training loss, its test loss can continue to reduce. Perhaps because once it finds a class of parameters that can fit all the data it can then find one that does so with the lowest complexity. 🤯

    4 korisnika označavaju da im se sviđa
    Poništi
  21. proslijedio/la je Tweet
    14. sij

    An incoherent theory. If intelligence had no function but to persuade, there's no obvious reason for us to evolve to be persuaded - *at all* - or for the resulting pattern of speech noises randomly also working to build spaceships. This is mere cynicism.

    Mindbending thought. "IQ developed as a byproduct of our social intelligence, and evolved to persuade others and justify our own actions." H/T 's newsletter cc
    14 replies 7 proslijeđenih tweetova 123 korisnika označavaju da im se sviđa
    Poništi

@NPCollapse još nije objavio nijedan Tweet.

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:

    ·