Cool stuff. Looking forward to more realistic NPCs and player decision driven stories in an open world sandbox
GAN Theft Auto [video]
21–30 of 99 posts
Re: GAN Theft Auto [video]
#22Great work! Hacker News still seems to have a deeply skeptical culture with regard to machine learning - not sure why. There's always someone saying it's "not novel" and it's "just doing x". Overfitting is a known issues in machine learning, people. If you still think all neural networks are doing is memorizing the dataset completely in the year 2021 - you might want to revisit the topic. It is one of the first conce…
> Hacker News still seems to have a deeply skeptical culture with regard to machine learning
Is... that a bad thing? Skepticism is good. When it's about something as hyped as "deep learning", even more so.
Re: GAN Theft Auto [video]
#23Impressive. Makes you wonder if at some point in the future there isn't a game engine any more but tons of training material and you play in a generated dream.
Certainly impressive. And sure, maybe in a distant future. Though I think this is like one of those things where creating a working prototype that is 75% complete is the "easy" part. The other 25% (which you need for an actual working product) will take forever. Like self driving cards, nuclear fusion, etc.
Text dungeon was very much a 75% complete thing, yet it's greatly entertaining in its own right. I would happily play a dream game which just falls apart sometimes.
Re: GAN Theft Auto [video]
#24It should be possible to take arbitrary video training data (whether from a game or real-life) and automatically reconstruct the 3D models of all vehicles in the scene (and the skybox) and "play back" the scene in a video game engine.
This is the direction virtual and augmented reality is headed (Facebook Codec Avatar, and their room reconstruction technology).
Re: GAN Theft Auto [video]
#25One throwaway line about GAN operating systems now made me want to see a shell GAN. Keypresses as inputs, 80x24 terminal screens as outputs. Could a neural network dream of Unix?
Re: GAN Theft Auto [video]
#26I like your YouTube videos in general and think this content is a great benefit to the community.
I wouldn’t take the few negative comments personally - I’ve seen many GAN architectures that heavily overfit (including my own bobross pix2pix) get a lot of praise, while ‘less violating’ models (like yours) get more skepticism. Skepticism isn’t bad! But I’d wager in your case it may be because you’re a YouTuber, and other ml YouTubers are notorious for ripping off content (eg Siraj).
Not really related to this, but I’d personally love to see the difference in training times it would take an RL agent to adequately learn to drive a car in gta versus adequately flying a helicopter.
Re: GAN Theft Auto [video]
#27GPUs: am I a joke to you? Instead of using them to render polygons, let’s use them to train neural networks that produce models that make them unnecessary. I’m oversimplifying - but pretty wild nonetheless.
Re: GAN Theft Auto [video]
#28Great work! Hacker News still seems to have a deeply skeptical culture with regard to machine learning - not sure why. There's always someone saying it's "not novel" and it's "just doing x". Overfitting is a known issues in machine learning, people. If you still think all neural networks are doing is memorizing the dataset completely in the year 2021 - you might want to revisit the topic. It is one of the first conce…
> Hacker News still seems to have a deeply skeptical culture with regard to machine learning Is... that a bad thing? Skepticism is good. When it's about something as hyped as "deep learning", even more so.
>Is... that a bad thing?
Yes, when it is there for no valid reason, or ridiculous reasons. Skepticism is not a default position you can take like a toddler refusing to eat their vegetables. You need some informed (and non-fallacious) intelligent reasoning behind that. "I'm skeptic about this thing using X because X is so hyped these days" is not such reasoning.
Re: GAN Theft Auto [video]
#29Great work! Hacker News still seems to have a deeply skeptical culture with regard to machine learning - not sure why. There's always someone saying it's "not novel" and it's "just doing x". Overfitting is a known issues in machine learning, people. If you still think all neural networks are doing is memorizing the dataset completely in the year 2021 - you might want to revisit the topic. It is one of the first conce…
> Hacker News still seems to have a deeply skeptical culture with regard to machine learning Is... that a bad thing? Skepticism is good. When it's about something as hyped as "deep learning", even more so.
Skepticism is good when it targets bold claims with vague proof. This is not a bold claim (it's a video demo showing the process) and its proof is not vague (you can inspect the source). Skepticism over something like GPT-2 without more than sample output is good. Skepticism over GPT-2 with a workable demo and source is unhelpful.
Re: GAN Theft Auto [video]
#30I like your YouTube videos in general and think this content is a great benefit to the community. I wouldn’t take the few negative comments personally - I’ve seen many GAN architectures that heavily overfit (including my own bobross pix2pix) get a lot of praise, while ‘less violating’ models (like yours) get more skepticism. Skepticism isn’t bad! But I’d wager in your case it may be because you’re a YouTuber, and oth…
Because you're using the word "your" I just feel the need to clarify that I didn't create this. I just saw it on YouTube and thought it was neat.