AIs can play video games "with a terrifying level of aggression and precision", but are VERY BAD at longer-term strategy
Ayeee, AIs are terrifying in video games that center on short burst interactions- I think Dota 2 was the one where it beat out a human team? It had to be fed some insane amount of data (4k years of gameplay vs itself?) but I think once you gave it something like Starcraft, the AI failed badly. Prob cause there's no clear optimal path, and like they said, a lot of long term planning you have to adjust on the fly?
the AI will get stuck in local extrema before finding the globabl one. So there are more complex methods that force it to go explore more of the search space.
Ah yes, the bane of my entire field. I guess it's some solidarity that AIs get to suffer the same as me XD
AIs are also really good at discovering and exploiting shortcuts and glitches in the simulation to cheat their way through
So there was a funny/interesting instance a little bit ago (2019? I'll have to dig for it cause it was interesting to show how literal AI are) where a team thought they got their AI to go from satellite maps to street maps and vice versa. But it turned out the AI was being told to remake an 'accurate' map, it'd code the image as color variations in the image. Which no human could probably see, but it'd just pull that to remake the copy. I think they finally caught on cause the image file sizes were a lot bigger than they were supposed to be XD
But ohh, I'm curious to see what you think of Gideon of the Ninth! I think I have the ebook somewhere languishing on my hard disk, but space necromancers is an interesting premise? But with Buffyspeak 21st century lingo?
I think Dota 2 was the one where it beat out a human team?
Yeah, I think that was the one mentioned in the book, too, and it was something crazy like that in terms of years of practice against itself -- maybe hundreds of years rather than thousands, but definitely longer than a human lifetime XD
where a team thought they got their AI to go from satellite maps to street maps and vice versa. But it turned out the AI was being told to remake an 'accurate' map, it'd code the image as color variations in the image. Which no human could probably see
Yes! There was a reference to that incident as well, in the book, in the "lazy cheater AIs" chapter. It's another great example of "wow, that's a really simple solution to... totally not the problem we want you to solve" XD
Ah yes, the bane of my entire field. I guess it's some solidarity that AIs get to suffer the same as me XD
Hehe, yeah, not wanting to search for global extrema felt really relatable! (In general the book does a really nice job of, well, not humanizing AI, since that's like the opposite of the intent, but making it fun to read about. The neural networks described do feel like characters, even though the book is careful to show they're not, if that makes any sense.
space necromancers is an interesting premise? But with Buffyspeak 21st century lingo?
Space necromancers is not really my #aesthetic, but the book sure does commit to it hard... XD Not in a bad way -- it does some clever things that I can appreciate even though it's not really my thing. I was expecting the language (more Tumblr memes than Buffy, as I understand it; I think I don't know nearly enough Tumblr memes to catch them all, which is probably a good thing) to bother me, because it's not clear how someone who grew up in those circumstances would know all the things the book is referencing, and that part's true, but I'm not reading it as a close third POV, I'm just reading it as a 21st century voice narrating the story and only letting us see one character's thoughts, and that's working for me so far.
It's another great example of "wow, that's a really simple solution to... totally not the problem we want you to solve" XD
It's weird cause a lot of the networks get such a huge degree of freedom, they'll just look at a problem and go 'hmmm, no rules??? No rules?! GO!' XD Which goes to show computers do exactly what they're told- nothing else, and nothing more.
The neural networks described do feel like characters, even though the book is careful to show they're not, if that makes any sense
Yeah, I guess de-mystifying them by 'characterizing' them as quirky things like Office Assistant? Cause normal pop culture has AI as these complex, alien black box behemoths, which...they're not? Like we're prob centuries away from 'sentient' AI, but we certainly have quirky weird AI.
those circumstances would know all the things the book is referencing, and that part's true, but I'm not reading it as a close third POV, I'm just reading it as a 21st century voice narrating the story and only letting us see one character's thoughts, and that's working for me so far.
Huh. I guess if Tumblr memes were the in universe surviving relic of our age a la Shakespeare (which would be...bizarre to think about trying to reconstruct) but I like your take on it! Kinda LOTR like I suppose, but by tumblr?
Ayeee, AIs are terrifying in video games that center on short burst interactions- I think Dota 2 was the one where it beat out a human team? It had to be fed some insane amount of data (4k years of gameplay vs itself?) but I think once you gave it something like Starcraft, the AI failed badly. Prob cause there's no clear optimal path, and like they said, a lot of long term planning you have to adjust on the fly?
the AI will get stuck in local extrema before finding the globabl one. So there are more complex methods that force it to go explore more of the search space.
Ah yes, the bane of my entire field. I guess it's some solidarity that AIs get to suffer the same as me XD
AIs are also really good at discovering and exploiting shortcuts and glitches in the simulation to cheat their way through
So there was a funny/interesting instance a little bit ago (2019? I'll have to dig for it cause it was interesting to show how literal AI are) where a team thought they got their AI to go from satellite maps to street maps and vice versa. But it turned out the AI was being told to remake an 'accurate' map, it'd code the image as color variations in the image. Which no human could probably see, but it'd just pull that to remake the copy. I think they finally caught on cause the image file sizes were a lot bigger than they were supposed to be XD
But ohh, I'm curious to see what you think of Gideon of the Ninth! I think I have the ebook somewhere languishing on my hard disk, but space necromancers is an interesting premise? But with Buffyspeak 21st century lingo?
Reply
Yeah, I think that was the one mentioned in the book, too, and it was something crazy like that in terms of years of practice against itself -- maybe hundreds of years rather than thousands, but definitely longer than a human lifetime XD
where a team thought they got their AI to go from satellite maps to street maps and vice versa. But it turned out the AI was being told to remake an 'accurate' map, it'd code the image as color variations in the image. Which no human could probably see
Yes! There was a reference to that incident as well, in the book, in the "lazy cheater AIs" chapter. It's another great example of "wow, that's a really simple solution to... totally not the problem we want you to solve" XD
Ah yes, the bane of my entire field. I guess it's some solidarity that AIs get to suffer the same as me XD
Hehe, yeah, not wanting to search for global extrema felt really relatable! (In general the book does a really nice job of, well, not humanizing AI, since that's like the opposite of the intent, but making it fun to read about. The neural networks described do feel like characters, even though the book is careful to show they're not, if that makes any sense.
space necromancers is an interesting premise? But with Buffyspeak 21st century lingo?
Space necromancers is not really my #aesthetic, but the book sure does commit to it hard... XD Not in a bad way -- it does some clever things that I can appreciate even though it's not really my thing. I was expecting the language (more Tumblr memes than Buffy, as I understand it; I think I don't know nearly enough Tumblr memes to catch them all, which is probably a good thing) to bother me, because it's not clear how someone who grew up in those circumstances would know all the things the book is referencing, and that part's true, but I'm not reading it as a close third POV, I'm just reading it as a 21st century voice narrating the story and only letting us see one character's thoughts, and that's working for me so far.
Reply
It's weird cause a lot of the networks get such a huge degree of freedom, they'll just look at a problem and go 'hmmm, no rules??? No rules?! GO!' XD Which goes to show computers do exactly what they're told- nothing else, and nothing more.
The neural networks described do feel like characters, even though the book is careful to show they're not, if that makes any sense
Yeah, I guess de-mystifying them by 'characterizing' them as quirky things like Office Assistant? Cause normal pop culture has AI as these complex, alien black box behemoths, which...they're not? Like we're prob centuries away from 'sentient' AI, but we certainly have quirky weird AI.
those circumstances would know all the things the book is referencing, and that part's true, but I'm not reading it as a close third POV, I'm just reading it as a 21st century voice narrating the story and only letting us see one character's thoughts, and that's working for me so far.
Huh. I guess if Tumblr memes were the in universe surviving relic of our age a la Shakespeare (which would be...bizarre to think about trying to reconstruct) but I like your take on it! Kinda LOTR like I suppose, but by tumblr?
Reply
Leave a comment