Richard Ngo

@RichardMCNgo

AI safety, ML, philosophy. Research engineer . Searching for utopia.

London
Vrijeme pridruživanja: listopad 2009.

Tweetovi

Blokirali ste korisnika/cu @RichardMCNgo

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

  1. Prikvačeni tweet
    2. sij

    If we want people to work towards an amazing future, we need to construct visions of what that future might look like. Which motivated my latest (and longest) blog post, Characterising Utopia.

    Poništi
  2. 26. sij

    If having having children was (financially) net positive in pre-industrial societies, and it's net negative today, there must have been a crossover point where it was net neutral. I have no idea when, though - any guesses?

    Poništi
  3. 21. sij

    It's bizarre that I never realised that Kyoto and Tokyo have the same two syllables, just swapped.

    Poništi
  4. 18. sij

    Twitter, what are the best introductions to climate science? I'm surprised I haven't heard of any bestselling pop sci books about the field yet.

    Poništi
  5. 17. sij

    This point will become increasingly relevant as deep RL improves. See for instance: "We demonstrate empirically that an entirely model-free approach... can learn to to exhibit many of the characteristics typically associated with a model based planner."

    Prikaži ovu nit
    Poništi
  6. 17. sij

    I've never liked the distinction between model-based and model-free agents in deep RL. "Model-free" agents can still have knowledge about their environments buried in their neural networks (like humans have). Instead, how about distinguishing explicit vs implicit models?

    Prikaži ovu nit
    Poništi
  7. 17. sij

    In yet another shocking display of poor data security, it turns out that you can make a human start reciting confidential information just by putting them in sleep mode. Totally bypasses standard authentication procedures. Whoever let this feature ship should be fired.

    Poništi
  8. 16. sij

    Where descriptions are discrete sequences of items from a finite alphabet, giving rise to infinitely many possible hypotheses. Yet it's hard to imagine a reasonable alternative. Credit to this blog post for reminding me of the argument:

    Prikaži ovu nit
    Poništi
  9. 16. sij

    Occam's razor (that simpler hypotheses are more likely, all else equal) is not an assumption, it's a mathematical fact that follows from there being many more long hypotheses than short hypotheses. The real assumption is that we should measure simplicity by description length,

    Prikaži ovu nit
    Poništi
  10. 10. sij

    Today in my journey to becoming a true tech bro: had my first scooter meeting. The new wheeling and dealing.

    Poništi
  11. 10. sij
    Prikaži ovu nit
    Poništi
  12. 10. sij

    This tweet originally inspired by my musing that in a big world, there'd be far too many incredible web serials running in parallel to read. Yet in fact, my favourite (which has just resumed after a 2 month hiatus) stands far above the rest. Check it out!

    Prikaži ovu nit
    Poništi
  13. proslijedio/la je Tweet

    Love this from Marc Andreesen: "The world is a very malleable place. If you know what you want, and you go for it with maximum energy and drive and passion, the world will often reconfigure itself around you much more quickly and easily than you would think." —

    Prikaži ovu nit
    Poništi
  14. 5. sij

    People reveal so much information about what they really think and want by the jokes they make. The topics, the intonations, whose laughter they look for... It's like a mask suddenly dropping.

    Poništi
  15. 5. sij

    Today I've been brainstorming ways I wish I'd been taught each academic subject. Most of them probably wouldn't have worked - but I figure the longer I iterate on this, the more prepared I'll be for teaching my own kids. So how do you wish you'd been taught?

    Prikaži ovu nit
    Poništi
  16. 4. sij

    I'm thinking about doing a literature review of fictional depictions of immortality. What are your favourite examples? Pointers to previous work on this are also very welcome.

    Poništi
  17. 4. sij

    Main downside is lack of an app, although it's very new so I assume that's on its way. Plus a few small bugs + missing features. Here's a list.

    Prikaži ovu nit
    Poništi
  18. 4. sij

    I've only been using for a couple of weeks, but so far so amazing! It's gradually replacing a totally unstructured mess of hundreds of Google docs and evernotes. And I don't need to worry about where notes "really belong", because I can link anything from anywhere.

    Prikaži ovu nit
    Poništi
  19. 4. sij

    What non-AI research might contribute to technical AI safety? Here are 30 ideas; and I'd be keen to hear more!

    Poništi
  20. 3. sij

    As human knowledge grows, refactoring what's already known to help people reach the frontier more quickly becomes arbitrarily important.

    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:

    ·