Maker Of Errors

@MakerOfErrors

AVOID, THIS.

Milkdromeda
Vrijeme pridruživanja: listopad 2017.

Tweetovi

Blokirali ste korisnika/cu @MakerOfErrors

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

  1. Prikvačeni tweet
    9. stu 2017.

    To fix media, don't grade on truthfulness or prediction accuracy. They'll just sensationalize without technically lying. Instead, compare *reader's* predictive power against those who read a placebo article. In other words, isn't the problem; it's the solution.

    Poništi
  2. 5. svi 2019.

    When people say something abstract can't be reduced to a mathematical model, it's because *they* can't model it, not because there exists even a single thing which operates outside causality.

    Poništi
  3. 2. svi 2019.

    It's usually not by adding more and more detail to your model. (Although that might reveal cracks you can exploit.) If the 1st-order approximation wasn't good, adding 2nd order effects can't help much. You need a new paradigm to implement. You need an easier problem to solve.

    Prikaži ovu nit
    Poništi
  4. 2. svi 2019.

    Have you tried realizing you're solving the wrong problem? I believe that if you could only do the math enough times, you can make the predictions turn out better.

    Prikaži ovu nit
    Poništi
  5. 12. pro 2018.

    "The equation works in this context but not this one" should turn into a qualitative debate on why. That's where hypotheses come from. The process of hypothesis generation is a key, but neglected, art. Yet, it's not well understood, so not taught, and so most hypotheses suck.

    Prikaži ovu nit
    Poništi
  6. 12. pro 2018.

    Physicists are especially bad about this, especially with quantum mechanics. "I don't understand" should never be met with "you'll get used to it", but that's Formalism's answer. The Copenhagen Interpretation was a map/territory error.

    Prikaži ovu nit
    Poništi
  7. 12. pro 2018.

    "The fire in the equations" is just an API. It's a dynamic pattern-matching thing brains do when doing word problems. It's highly contextual, because as Wittgenstein pointed out, all language is. So, why do scientists retreat to Formalism, as a philosophy of mathematics?

    Prikaži ovu nit
    Poništi
  8. 3. pro 2018.

    More real than reality. Now, you have movie sound effects like you'd expect rather than real sounds. And space scenes that look less fake than the moon landings because they light it from multiple angles. But Goodhart's Law will make sims with Aristotle's, not Newton's physics.

    Tweet je nedostupan.
    Poništi
  9. 1. pro 2018.

    "At most א good people can exist." Calculating this upper bound Aleph is left as an exercise for the reader. So is increasing it to Aleph-null.

    Prikaži ovu nit
    Poništi
  10. 1. pro 2018.

    Most humans are indifferent to suffering, and usually won't go out of their way to cause or prevent it. It would only take a small fraction of global GDP to eliminate must suffering. Nothing's stopping you. If even a tiny number were truly benevolent, suffering would not exist.

    Prikaži ovu nit
    Poništi
  11. 1. pro 2018.

    Theodicy paradox: 1) benevolent, all-powerful god exists 2) unjust suffering exists ∴ ¬1 and/or ¬2 (proof by contradiction) Anthropodicy paradox: 1) Benevolent humans exist 2) unjust suffering exists ∴ ¬1 and/or ¬2 (proof by contradiction) Become good. You aren't. EA isn't.

    Prikaži ovu nit
    Poništi
  12. 24. stu 2018.

    "Ha, look! The chatterbot's lying about itself, and doesn't know I'm the one who told it to pretend to be a tech support worker. Isn't that funny?" "Nuke it from orbit. It's the only way to be sure."

    Prikaži ovu nit
    Poništi
  13. 24. stu 2018.

    The first AGI will be cute, right before it seizes the light cone. "Awww" is the human reaction to recognizing agency, when it's not threatening. Kittens pouncing is cute, but tigers aren't. But, kittens don't self-modify into tigers overnight. So, our intuitions are off here.

    Prikaži ovu nit
    Poništi
  14. 24. stu 2018.

    It also illustrates that a lot of Sensemaking is just post-hoc rationalization. This makes memories shift. When connecting the dots, be careful to minimize your biases. Don't connect unconnected things. Don't become a solar deity. Don't remake my errors.

    Prikaži ovu nit
    Poništi
  15. 24. stu 2018.

    This is part of a consolidation process, because I noticed my System 1 didn't grok Social Intuitionism. So, I translated it into social intuitions, building a character. I now have something vivid on which to attach the relevant hooks to related concepts.

    Prikaži ovu nit
    Poništi
  16. 24. stu 2018.

    The mind which acts identically to yours, except that it doesn't bother to rationalize decisions after Action Selection. Ask why it did something, and it always answers "I don't know". Or it gives an instrumental reason, but when asked why do that teminal goal, it doesn't know.

    Prikaži ovu nit
    Poništi
  17. 24. stu 2018.

    For example, the concepts "consolidating one's ontology" and "sensemaking" are mostly the same. Maybe sensemaking is a special case, or a single step in unifying one's ontology. I must think on this.

    Prikaži ovu nit
    Poništi
  18. 24. stu 2018.

    Consolidate your ontology. What's a special case of what, and can you generalize a principle to combine multiple concepts? Make explicit your rules for when to use which model. Are there cases not covered, or where multiple models apply? Which of your beliefs might conflict?

    Prikaži ovu nit
    Poništi
  19. 20. stu 2018.

    How do you know you haven't unwittingly worked on such a system before? If you've ever done troubleshooting, it would explain a lot.

    Prikaži ovu nit
    Poništi
  20. 20. stu 2018.

    The number of Nobel prizes that would need to be won just to understand such a motherboard. Or to manufacture one. Or to troubleshoot one. It'd be like Neuroscience if nature was actively malicious in her design process, rather than merely messy.

    Prikaži ovu nit
    Poništi
  21. 20. stu 2018.

    Make most transistors appear purposeful, but actually do nothing but scramble the true signals, so they can't be isolated without breaking the whole system. Instead, perform computations using the unexplained differences between best models of components, and actual behavior.

    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:

    ·