4 ms·

I'm waiting for the time where all this neural technology will start to be used for something good, like making better games with more NPCs who can really talk.

I'm waiting for the time where all this neural technology will start to be used for something good, like making better games with more NPCs who can really talk. Maybe never. The life-cycle seems to be research -> malicious use -> hipster nonsense -> memes -> abandonment. Edit: the problems with these images look very much like application of anisotropic smoothing. G'MIC has filters like that. You can make this stuff look more realistic by blurring it (gaussian) and adding noise (uniform). Blurring hides small-scale irregularities, while noise makes blurring less obvious by adding small-scale "grains" that you perceive as detail/texture.

ddeokbokki 8y ago
Brb, going to train a GAN with all my selfies so it can generate new selfies.
donaldknuth123 8y ago
>like making better games with more NPCs who can really talk Are no Animation companies or Game Devs trying that out? Seems like a fairly obvious use case but maybe they don't have the "expertise".
gwern 8y ago
Things do get applied. If you want applications of GANs to better games, did you notice the past few weeks discussion of using ESRGAN & other superresolution GANs to upscale artwork of old games to make them far prettier and highres?
- gambler 8y ago
  I was thinking more along the lines of an RPG with generated town population. Indie devs usually don't have resources to draw hundreds of portraits and record tends of hours of dialog, but if they could use ML to generate portraits and synthesize decent speech, it would allow them to create much bigger, more immersive worlds. Also, this stuff could be used to take a hand-drawn portrait and animate it with different expressions without rework.
  - gwern 8y ago
    "The future is already here, it's just unevenly distributed." Stuff like that will come, eventually. Like any cutting-edge R&D, there's a long valley of death from a lab demo to a globalized real-world product. It's a lot of work to package things up so reliably & cleanly that harried game devs can easily & usefully incorporate them into games.
    - gambler 8y ago
      >Stuff like that will come, eventually. Well, I'm not so sure. On one hand, games use graphics techniques that were developed couple years prior to their development. On the other hand, they fail to capitalize even on relevant AI research of the 60s and 80s. When they do, it looks amazing (e.g. FEAR AI), but it's very rare.
      - gwern 8y ago
        The excuse I always hear is that those techniques might be smarter but not as fun. That doesn't appear near as applicable to graphics stuff, where game developers have always been eager to apply and extend the cutting edge.
        crooked-v 8y ago
        The FEAR AI is actually simpler than you might think, though with some really good insights in term of applying behavioral design to gameplay. https://alumni.media.mit.edu/~jorkin/gdc2006_orkin_jeff_fear.pdf https://alumni.media.mit.edu/~jorkin/gdc2006_orkin_jeff_fear... > As much as we like to pat ourselves on the back, and talk about how smart our A.I. are, the reality is that all A.I. ever do is move around and play animations! Think about it. An A.I. going for cover is just moving to some position, and then playing a duck or lean animation. An A.I. attacking just loops a firing animation. Sure there are some implementation details; we assume the animation system has key frames which may have embedded messages that tell the audio system to play a footstep sound, or the weapon system to start and stop firing, but as far as the A.I.’s decision-making is concerned, he is just moving around or playing an animation. > Now let’s look at our complex behaviors. The truth is, we actually did not have any complex squad behaviors at all in F.E.A.R. Dynamic situations emerge out of the interplay between the squad level decision making, and the individual A.I.’s decision making, and often create the illusion of more complex squad behavior than what actually exists! > Imagine we have a situation similar to what we saw earlier, where the player has invalidated one of the A.I.’s cover positions, and a squad behavior orders the A.I. to move to the valid cover position. If there is some obstacle in the level, like a solid wall, the A.I. may take a back route and resurface on the player’s side. It appears that the A.I. is flanking, but in fact this is just a side effect of moving to the only available valid cover he is aware of.

ddeokbokki 8y ago

Brb, going to train a GAN with all my selfies so it can generate new selfies.

donaldknuth123 8y ago

>like making better games with more NPCs who can really talk Are no Animation companies or Game Devs trying that out? Seems like a fairly obvious use case but maybe they don't have the "expertise".

gwern 8y ago

Things do get applied. If you want applications of GANs to better games, did you notice the past few weeks discussion of using ESRGAN & other superresolution GANs to upscale artwork of old games to make them far prettier and highres?

gambler 8y ago
I was thinking more along the lines of an RPG with generated town population. Indie devs usually don't have resources to draw hundreds of portraits and record tends of hours of dialog, but if they could use ML to generate portraits and synthesize decent speech, it would allow them to create much bigger, more immersive worlds. Also, this stuff could be used to take a hand-drawn portrait and animate it with different expressions without rework.
- gwern 8y ago
  "The future is already here, it's just unevenly distributed." Stuff like that will come, eventually. Like any cutting-edge R&D, there's a long valley of death from a lab demo to a globalized real-world product. It's a lot of work to package things up so reliably & cleanly that harried game devs can easily & usefully incorporate them into games.
  - gambler 8y ago
    >Stuff like that will come, eventually. Well, I'm not so sure. On one hand, games use graphics techniques that were developed couple years prior to their development. On the other hand, they fail to capitalize even on relevant AI research of the 60s and 80s. When they do, it looks amazing (e.g. FEAR AI), but it's very rare.
    - gwern 8y ago
      The excuse I always hear is that those techniques might be smarter but not as fun. That doesn't appear near as applicable to graphics stuff, where game developers have always been eager to apply and extend the cutting edge.
      - crooked-v 8y ago
        The FEAR AI is actually simpler than you might think, though with some really good insights in term of applying behavioral design to gameplay. https://alumni.media.mit.edu/~jorkin/gdc2006_orkin_jeff_fear.pdf https://alumni.media.mit.edu/~jorkin/gdc2006_orkin_jeff_fear... > As much as we like to pat ourselves on the back, and talk about how smart our A.I. are, the reality is that all A.I. ever do is move around and play animations! Think about it. An A.I. going for cover is just moving to some position, and then playing a duck or lean animation. An A.I. attacking just loops a firing animation. Sure there are some implementation details; we assume the animation system has key frames which may have embedded messages that tell the audio system to play a footstep sound, or the weapon system to start and stop firing, but as far as the A.I.’s decision-making is concerned, he is just moving around or playing an animation. > Now let’s look at our complex behaviors. The truth is, we actually did not have any complex squad behaviors at all in F.E.A.R. Dynamic situations emerge out of the interplay between the squad level decision making, and the individual A.I.’s decision making, and often create the illusion of more complex squad behavior than what actually exists! > Imagine we have a situation similar to what we saw earlier, where the player has invalidated one of the A.I.’s cover positions, and a squad behavior orders the A.I. to move to the valid cover position. If there is some obstacle in the level, like a solid wall, the A.I. may take a back route and resurface on the player’s side. It appears that the A.I. is flanking, but in fact this is just a side effect of moving to the only available valid cover he is aware of.