EDIT 2: The RTA WR for BLJless Lonely mushroom run is actually 9.73 xcam / 9.63 grab by Atmpas: ruclips.net/video/iHFjBrlxHFc/видео.htmlsi=u9G8RT1Tw9Wq638W EDIT: The fastest crazy-box run is 10.20 xcam by TaylorSwiftSimp (10.06 grab): ruclips.net/video/qKcCoJjLFag/видео.html There is also a 10.23 xcam (~10.00 grab) by Harrison Schindele: ruclips.net/video/wkA1xSH7yjc/видео.html
Ah sorry, I was going off of the "ultimate star spreadsheet". I just found it and will add it to the description / make a pinned comment. Glad you enjoyed the video!
These videos are a highlight of my day when the come out. Thank you! Is the search actually completely random? Like, are there a lot of attempts where it just turns around and long jumps the wrong way and dies?
Yes it's completely random, and most of the time it will not find faster ways to get to a specific point in the level. The trick is that it only keeps track of runs that do, and keeps track of more or less all points in the level and how it got there (keeping only fastest run for each point).
@@Krithalith Ah! Then do you direct the search (manually? automatically?) to preference attempts that head (vaguely?) in the direction of the star for example? Using known-fast paths to points in that direction?
@@hrrld There is initially no preference for runs that get "close" to the star or whatever, all points in the level are treated equally. This does mean that a lot of time is spent on pointless areas of the level. Once a run to the star is found its blocks are selected with a higher weight, but other blocks are not excluded. There are some other ways to cleverly restrict the search space without introducing significant biases as well.
The last strat looked siiiiiiiiick! I love quick jumpkicks! Also, why does that crash happen? I saw ted mentioning null ceiling, but is there a quick step by step explanation you can give me if it's not gonna take too much time for you to write it down?
Would it be possible to set the playback camera to be like lakitu cam? I genuinely kind of have a headache now from watching this haha. In any case, excellent work as always!
I chose Mario cam for this one because lakitu hides a lot of the water sections and can get far behind during quick climbs. Mario cam is definitely a bit fast sometimes though :p
This makes me curious as to what other game crashes Scattershot can find if the AI can be made to search them out. As someone who has an unusual obsession with crashing games, I'd be interested to see a series on that!
SM64 optimization / route finding algorithms. No neural networks involved. In this video I show the results of some low-resolution searches, and I show off what was found, hopefully leading to something useful for RTA.
I'm curious as to how Scattershot knows about SM64 glitches; is it imparted with knowledge about them or does it just find them as part of its bruteforcing strategy?
EDIT 2: The RTA WR for BLJless Lonely mushroom run is actually 9.73 xcam / 9.63 grab by Atmpas:
ruclips.net/video/iHFjBrlxHFc/видео.htmlsi=u9G8RT1Tw9Wq638W
EDIT: The fastest crazy-box run is 10.20 xcam by TaylorSwiftSimp (10.06 grab):
ruclips.net/video/qKcCoJjLFag/видео.html
There is also a 10.23 xcam (~10.00 grab) by Harrison Schindele:
ruclips.net/video/wkA1xSH7yjc/видео.html
i dont mean to be THAT person, but i do have a 10.20 xcam with box ruclips.net/video/qKcCoJjLFag/видео.html
@@TransRightsOrKYS Ahh I see, thanks! Just updated the info.
i am in love with how creative (funny word choice) these movement patterns are. it's so free and absurd in a really beautiful way.
ok dude what the actual floop
holy moly is that THE manama3141ted5926
I love these! Some of those were insane lol
Yo wassup kosmic
The GWK ascent is too powerful to beat but geez those other strats are insane :O Awesome stuff as always!
Scattershot never fails to crush my expectations and leave me speechless :D low-res search is crazy, I love it
5:48 was nasty. GJ!
Great video as always! Btw I believe blj-less box record is 10.23 xcam by me on Ukikipedia. Still no match for scattershot!
Ah sorry, I was going off of the "ultimate star spreadsheet". I just found it and will add it to the description / make a pinned comment. Glad you enjoyed the video!
2:35 ok that slidekick is so fricking sick
3:34 was a kreygasm and a half
Next BLJless full TAS might be insane
This is crazy man 😂
These videos are a highlight of my day when the come out. Thank you!
Is the search actually completely random? Like, are there a lot of attempts where it just turns around and long jumps the wrong way and dies?
Yes it's completely random, and most of the time it will not find faster ways to get to a specific point in the level. The trick is that it only keeps track of runs that do, and keeps track of more or less all points in the level and how it got there (keeping only fastest run for each point).
@@Krithalith Ah! Then do you direct the search (manually? automatically?) to preference attempts that head (vaguely?) in the direction of the star for example? Using known-fast paths to points in that direction?
@@hrrld There is initially no preference for runs that get "close" to the star or whatever, all points in the level are treated equally. This does mean that a lot of time is spent on pointless areas of the level. Once a run to the star is found its blocks are selected with a higher weight, but other blocks are not excluded. There are some other ways to cleverly restrict the search space without introducing significant biases as well.
@@Krithalith big approve 🙇 - cannot wait to see what you (and it) come up with next!
The last strat looked siiiiiiiiick! I love quick jumpkicks! Also, why does that crash happen? I saw ted mentioning null ceiling, but is there a quick step by step explanation you can give me if it's not gonna take too much time for you to write it down?
very cool
erm what the sigma
Lmao nice null ceiling crash, is it worth guarding against crashing cases?
It's necessary to not have the search crash lol. One of the first things I did was unload the moles cause I don't trust them anyway.
Would it be possible to set the playback camera to be like lakitu cam? I genuinely kind of have a headache now from watching this haha. In any case, excellent work as always!
I chose Mario cam for this one because lakitu hides a lot of the water sections and can get far behind during quick climbs. Mario cam is definitely a bit fast sometimes though :p
wow so fast :o
This makes me curious as to what other game crashes Scattershot can find if the AI can be made to search them out. As someone who has an unusual obsession with crashing games, I'd be interested to see a series on that!
How would you detect how close it is to a crash?
you can crash the game in 0s by jiggling the cartridge
not sure what the fastest known crash is by legal inputs only - some kind of PU in overworld?
Fastest crash is file select crash
@@snared_ null ceiling in bob would be the fastest crash
fastest softlock is lakitu w/ 0 hp in jp
this just showed up on my reccs whats going on? is this a neural network/evolutionary method of tasing or smth?
SM64 optimization / route finding algorithms. No neural networks involved. In this video I show the results of some low-resolution searches, and I show off what was found, hopefully leading to something useful for RTA.
what is scattershot what does low-res entail why is he scaling the mountain and why is none of this explained
because he doesn't want to explain it every video
edit: my joke radar might have malfunction
Forgive my ignorance, but what's x-cam timing?
Could scattershot be used for ABC?
18h ago
I'm curious as to how Scattershot knows about SM64 glitches; is it imparted with knowledge about them or does it just find them as part of its bruteforcing strategy?
Scattershot has no knowledge built in, all glitch finds are accidental.
ok wtaf