-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Metal Gear Solid AI Video Hallucinations
- Thread starter EviLore
- Start date
adamsapple
Or is it just one of Phil's balls in my throat?
ManaByte
Member
Even the AI was fooled
That scene isn't included in the recent collection, is it? Someone will cancel Konami over it.
deeptech
Member
The deepest forum post in the world, this here...
Also the female Raiden looks so cool in that little part, if they ever do a remake or movie, I would love it instead, it somehow fits given he doesn't have a "real" identity, not even he is sure who or what he is...
Last edited:
Yes, it decides what to do based on its training data (a very large collection of video files, in this case).Real nightmare fuel.
Also, does the AI decide on its own what to do?
If so, I at least consider it impressive that it attempted to do the pose with Liquid raising his arm.
Grildon Tundy
Member
AI has manifested levels of schlubiness mankind could only dream of.
Holammer
Member
Eventually I expect games to get bespoke AI models specifically trained for them (keeps 'em lightweight) and integrated so they don't hallucinate.
It'll know how Snake looks like and never miss a beat in rendering him correctly. Question is when we get a GPU capable of doing it, I think the first use of it will be stuff like water simulation, clouds and skyboxes.
It'll know how Snake looks like and never miss a beat in rendering him correctly. Question is when we get a GPU capable of doing it, I think the first use of it will be stuff like water simulation, clouds and skyboxes.
Kuranghi
Gold Member
Dr Disrespect as Otacon there at 01:09
Ironic, given Otacon was in the other position in his controversy
Moses85
Member
AI has manifested levels of schlubiness mankind could only dream of.
Big Baller
Al Pachinko, Konami President
Grildon Tundy
Member
I'm confused. Is Edward Norton in that gif supposed to represent AI's reaction to pulling one over on humanity?
rabbity-thing
Member
I wonder if the AI thinks that non-AI-generated media looks just as weird as we think the AI-generated stuff looks.
Like, does the AI get mad that mother nature just can't get hands right?
Like, does the AI get mad that mother nature just can't get hands right?
Last edited:
Moses85
Member
No, snake looks like him.
Badlucktroll
Member
metaru geya!?!!!??
TGO
Hype Train conductor. Works harder than it steams.
It is.
cash_longfellow
Member
AI visualizes what I visualized the one time I accidentally had 4 hits of LSD back in the early 2000s, because a friend bumped my guys arm while he was dropping from the vial on my tongue……to add, this is not a good thing 
John Bilbo
Member
D̷o̷ ̵̦̕A̶͖͗n̷̲̿d̴̛̪ŗ̵͝o̷̜̾i̷̪͝d̶͓̑s̸̯̾ ̷̞̭̅̚D̷̜̅͋ͅr̷̳̃̊ė̷̞͛å̶͉̈́ͅm̷͉̾ ̵̪̝̓͒ǫ̸̛̹f̸̩̜̈ ̵̣̻͝Ė̵͖l̴͍̆ȩ̵̼̕c̸̨̞͊̀́t̸̺̭̎r̸̘̥̙̽i̷͎͗̈́͝c̵͔̦̗̏̃͝ ̴͍̬̎̓͗Š̴̢͇́h̷̘̑͛̚ę̴̹̌̊͛̕ē̴̥̠͇͔͐̕p̸̛̘̱͊͛̃ͅ?̷̛͚͊̊͑
Brigandier
Member
The ending 
diffusionx
Gold Member
Weird how the AI immediately interpreted Otacon as Todd Howard. The first Raiden is nightmare fuel. It seems to not be able to figure out how the PSX characters walk.
lol AI.
lol AI.
Last edited:
VulcanRaven
Member
They didn't remove anything.
Radical_3d
Member
That was Ocelot’s plan all along.
Even the AI was fooled
Jack Package
Member
Scott Bakula/Glen Howerton as Otacon is not something I'd have ever thought of...
Nor is young Bruce Campbell as the twin snakes, but now I want to see someone make that!
Also, just me or did Johnny/soldier at the end look like James MacAvoy?
Hudo
Gold Member
It depends on the training data (videos in this case). It is most likely some sort of Transformer-based network, so it has learned correlation between sequences (frames). Essentially, it is trained to generate matching (based on it's built-up latent space) next or missing frames in a sequence. Most Transformers are still auto-regressive, so they generate the next element in a sequence based on what has come before and use that as new input and just repeat this process. Obviously, SORA and other networks are a bit more sophisticated in their training process and probably have some post-processing going on. But most of the time, it's basically just brute-forcing quality with an enormous quantity of training data on large neural networks and not so much using really sophisticated architectures.Real nightmare fuel.
Also, does the AI decide on its own what to do?
If so, I at least consider it impressive that it attempted to do the pose with Liquid raising his fist.
TL;DR: It decides what the next frame should be based on previous frames in the input sequence and what it "knows", which is based what it has learnt (training data) and how it has learnt (loss function and training objective like sequence prediction or prediction of missing-element, which is a variation on sequence prediction, etc.)
RoadHazard
Gold Member
It's still ridiculous to me that AI can now do this, even if the results are often completely braindead.
NeoIkaruGAF
Gold Member
2:07 that independent arm is hilarious.
2:42 WTF there’s a dude behind Sniper Wolf’s body??
2:42 WTF there’s a dude behind Sniper Wolf’s body??
Corporal.Hicks
Member
It's hard to believe that AI could have done this.
Dirk Benedict
Gold Member
Took me out when they spontaneously morphed into Incredibles-like designs.
Power Pro
Member
I never knew I wanted Mel Gibson to play Ocelot so much