Eliezer

19 ms·

1. 181.▲
   by Eliezer 10y ago

   Or in a more intuitive sense, humans didn't seem to need brains 1000x as large to get one unit of improvement in practical effectiveness over chimpanzee cognition. But yes, this stuff is complicated, hence the paper not being two para
2. 182.▲
   by Eliezer 10y ago

   As it says in the paper, the evolution of hominids doesn't look like this; population genetics says that if hominid brains got larger, the marginal fitness returns on neurons went up.
3. 183.▲
   by Eliezer 10y ago

   I'm getting kind of sick of people who haven't read 2% of the stuff imagining what they think we've never talked about over the last 16 years. https://intelligence.org/files/IEM.pdf
4. 184.▲
   by Eliezer 10y ago

   This is excellent. If you want to see what real discussion of an AGI alignment issue looks like, please read this.
5. 185.▲
   by Eliezer 10y ago

   I'm reading this thinking, "Wait, why is Anders introducing this algorithm... Oh."
6. 186.▲
   by Eliezer 10y ago

   Except that you mainly need the massive compute for the training, not the commodity execution of the network!
7. 187.▲
   by Eliezer 10y ago

   Came here to say the same. It's not impossible, but you want to see multiple replications, and p a lot lower than 0.05, before believing study results like those.
8. 188.▲
   by Eliezer 10y ago

   This article pleases the Market Economics Fairy, who rarely sees anyone thinking top-to-bottom in supply-demand balances.
9. 189.▲
   by Eliezer 10y ago

   Will there ever be a VR option for those of us who still wear glasses?
10. 190.▲
    by Eliezer 10y ago

    And of course the last one in the sequence is "Goto Considered Harmful Considered Harmful Considered Harmful". Yet another illustration of Ultrafinite Recursion! ( https://m.facebook.com/story.php?story_fbid=1015
11. 191.▲
    by Eliezer 10y ago

    And that's how that guy whose face appeared a few times in ImageNet became the world's most wanted terrorist, on the run for thousands of crimes.
12. 192.▲
    by Eliezer 10y ago

    I'm sorry, but seeing the phrase "blockchain as a service" has been the final straw causing my sanity to snap. brb, murder spree
13. 193.▲
    by Eliezer 10y ago

    Some people get pointers immediately after one exposure before puberty. Not every student needs to practice. Some people, who maybe never came to your university at all, will never get linear algebra matter how much they practice. To say
14. 194.▲
    by Eliezer 10y ago

    > Ultimately I find this a bit ridiculous given how vocal some individuals in the tech field are about the 'dangers of Ai'. They harp and harp about what could happen if more capable Ai gets developed but you don't see the
15. 195.▲
    by Eliezer 10y ago

    (I am relatively excited about this research direction. It seems like the sort of thing that might lead to genuinely useful components of a safer AGI system later.)
16. 196.▲
    by Eliezer 10y ago

    I noticed that YouTube's recommendations had suddenly gotten better! I wondered if they were using a new statistical approach, or had just started really optimizing at all because the old recommendations were extremely naive. I'm
17. 197.▲
    by Eliezer 10y ago

    WHAT. That is one implausible SF premise that just allegedly came true there.
18. 198.▲
    by Eliezer 10y ago

    Somebody, one of these days, should design a cryptocurrency whose point is just to NOT get stolen.
19. 199.▲
    Movable Housing for Scalable Cities: A Reply to YC(steemit.com)

    27 points by Eliezer 10y ago|8 comments
20. 200.▲
    by Eliezer 10y ago

    I thought the argument was going to be "Conditionals are bad for running on GPUs."
21. 201.▲
    by Eliezer 10y ago

    Sorry, it took a while, but I wrote two Arbital nodes for you: - https://arbital.com/p/subjective_probability/ - https://arbital.com/p/likelihood_vs_pvalue/
22. 202.▲
    by Eliezer 10y ago

    This is an excellent analogy! "Ants can carry fifty times their own weight. Why would humans wipe out the ants on a construction site instead of using food rewards to train them to help remove dirt?" Answer: backhoes are faster a
23. 203.▲
    by Eliezer 10y ago

    Nice derail! But the fact that there isn't yet a 400-lightyear expanding hole in the Milky Way suggests that corporations are not, in fact, as dangerous as unaligned machine superintelligences are being alleged to be.
24. 204.▲
    by Eliezer 10y ago

    True at NASA and in many other places that spend 100x as much per line of code, albeit it's more like 95% surety.
25. 205.▲
    by Eliezer 10y ago

    The next generation of geniuses is growing up understanding why Friendly AI is hard. That's actually pretty damned encouraging.
26. 206.▲
    by Eliezer 10y ago

    "Tiling agents" would be the most arguably relevant. Provably correct agents that approve the construction of other probably correct agents obeying similar invariants.
27. 207.▲
    by Eliezer 10y ago

    Maybe the plot was sort of wuxia-ish? I haven't seen the movie so could be compleeeetely off base here, I was just trying to imagine what about the trailer I saw could map onto something that worked surprisingly well in China.
28. 208.▲
    by Eliezer 10y ago

    Previously, neural networks were trained by taking single steps down the direction of sharpest gradient for the network. However, in deep networks with lots of layers, the backprop algorithm (which I assume you already know about) got stuc
29. 209.▲
    by Eliezer 10y ago

    Obvious but maybe wrong answer: because it's not linear, if you trap nearly all the IR it stays until converted to visible, otherwise it still escapes after a couple of bounces.
30. 210.▲
    by Eliezer 10y ago

    It's not hard to imagine a future writer looking back and saying, "Of course Trump lost the election, he was no Kennedy or Obama and widely hated - a clear example of the rule!"