the good thing about OOT is that there are ALWAYS more optimizations and new glitches and applications of glitches to be found. so like that new route was only 30 (or 15) seconds faster, but yeah, since there are always new things to do, there's always more time to save, it'll definitely get to be a bigger difference from the old route in time. idk if there's another srm-sized glitch left in oot, but even if not, there are always some things here and there left to be discovered.
4:23 Speedrunner loses a hand but instead of a bionic hand their muscles get electrically linked to the inputs of a Gamecube controller and suddenly speedrunning is their superpower. ;)
A full Dijkstra algorithm on the node graph of OoT could do it, it would also take a long time to run and would be exhausting to represent OoT as a linked graph.
This is an interesting graph problem. A variant of the traveling salesman but going to certain nodes (getting an item) unlocks additional edges. I think it’s not completely unreasonable to setup the data structure. It vaguely reminds me of an old ACM programming contest question with moving submarines. This is still way harder than that, mind you. Would have to be greedy and wouldn’t give THE best answer but could give a pretty good one. I don’t know enough about the OoT speed run to just make it, though. Where do routing discussions happen? I may be biting off way more than I can chew but I’d like to take a stab at such an AI.
Rando already needs to generate a graph of what things depend on eachother and then walk it to determine if the seed is beatable. That seems like a good place to start.
To fair, AI is not that smart as people think it is. It would probably end up creating a impossible to beat route or a very bad one that actually waste more time instead of saving. Lol
oh yeah like it would be able to figure out some frame-perfect un-pause-buffered not-human-viable strats, but how the hell would an AI figure out something like text transfer glitch lol. maybe if you had an AI running OOT for two decades, the same amount of time that humans have been running it lol... but yeah when i was a kid and i first learned about speedrunning though, that is what i thought TASes were haha. i thought they were like a computer program beating a game in the most perfect way possible, not that they were speedrunners painstakingly recording each action over and over for hundreds of hours in order to get something perfect lol, advancing like a frame at a time for many tricks. i like the reality of TASes more lol cause damn that really is something
Honestly as an AI engineer, I’ve actually expressed a lot of interest in building speedrun and randomizer routes using AI and algorithms. I’d really love to work with OOT speedrunners and tech experts on this. I guess we wouldn’t have the AI finding glitches eg srm or text transfer, but we could characterize where these glitches are possible and create dependency graphs to determine the best routes using the tricks. That’s what routing is-not figuring out tech, but applying the tech in succession after we already understand it. Although some things might be computationally lofty (okay when we can let an optimizer run for days-we don’t need a new route every hour), a lot of the choices for a route we can tune to avoid impossible tricks or schedules unless they’re absolutely necessary. Not everything in AI is just a black box, AI’s are as smart as we build them to be, and I think computing optimal routes that are also RTA is definitely within feasibility. We would just need people to put the work in.
@@gwen6622 it would need to force it's way through the game and probably never figure out how the sandstorm after GTG works or stuff like slingshot and what stuff you can and can't shoot
@@MichaelSchwartzJazz I've always wondered where the AI parts come in since it seems like a lot of those dependencies and edge weights would need to be supplied by us, at which point it seems less like an AI problem and more like brute-forcing a travelling salesman problem.
I mean if someone wanted to we could just run a computer program to optimize a route. It would just take some time to find the time between every possibility. The idea would be to make a graph (like vertices with lines connecting them) and add the time to go from one point to the next as a weight. Then run a program which finds the optimal path. There may be some problems with conditionals, like if there's an action you can take with multiple possibilities but once you choose one you can do it again, but that could be sorted out fairly easily by just making a bunch of graphs with only one of those possibilities and optimizing them all, then choosing the best one of the bunch. The other difficulty is bombs, but idk there's probably a way.
Yeah, basically this would be a fancy traveling salesman problem. Edge weights based on items in inventory. That by itself wouldn’t be too hard I don’t think, though it would be tedious to setup and debug (in my experience with such things anyways). From there one could build on top with other fun routing tools, maybe.
@@SilkNeon What do you mean the traveling salesman problem isn't hard? The entire point of that problem is how exceptionally difficult it is for computers to solve it.
@@TheLastScoot I mean, it’s not too hard to get a pretty good answer. It’s hard to get the best answer. But we use greedy algorithms to approximate an acceptable answer in these all the time. When I say not hard I mean it more in the sense that it’s used as a setup to learn greedy algorithms in college. Meaning it could be of value in routing, I think.
@@SilkNeon Oh if it's just about getting a good enough one then fair enough, but then it becomes a question of whether that even ends up outperforming speedrunners coming up with routes or not. Sounds interesting at least.
@@SilkNeon you might be underestimating the complexity of this particular problem. as you note, finding _a_ solution is practical, well-trodden ground, even specific to OoT in some capacity wrt randomizer logic. optimizing that same problem for time is a wildly different beast. consider that the TTYD community has crowdsourced a solution to the recipe order for their 100% TAS using a highly parallel bruteforce graph search program. there are 57 recipes that can naively be completed in any order (pre-optimization, search space order 10^76), but it's ultimately just two consecutive trips to the same NPC, and the current leading candidate is just over a minute long. check out this video if you're unfamiliar and want more specifics: ua-cam.com/video/kpn30v5Y_YQ/v-deo.html despite the straightforward nature of the problem, the engineering effort to implement this solver was quite significant, and it ultimately faced unavoidable technical hurdles that impaired its overall effectiveness: for example, there exist more possible routes than atoms in the universe, and so there is no method for distribution of work, as a central server literally cannot retain information on the searched space indefinitely. comparatively, the graph of OoT -- even excluding SRM -- is incomprehensibly complex, with (at least) hundreds of additional nodes and variable edge lengths owing evolving item availability (including bomb drops!), the day/night cycle, etc. we can dream of a meta-OoT that could ingest the game's source to programmatically produce such a graph/model of the game world, but each and every added degree of freedom is another step toward the (present-day) impossible solely on a computational basis, too. it is true that we're not seeking TTYD's single-frame level of optimization, but it would still need to be reasonably precise: consider how incremental this new route wound up being relative to the time invested. I don't mean to totally kill your enthusiasm, but I intensely doubt it's feasible for OoT to have a wholly AI-generated 100% route on a multi-decadal timescale. that said, if you're into the concept, there are other speedgames out there that would for sure offer more fertile grounds for such an effort, and I would be very excited to see more of this style of approach within the speedrun community.
Quick question, if I put a bottle in Nayru's Love slot before getting NL, will I be locked out of getting NL? I seem to remember something about some items not being given if you already have something in the slot.
@@magmivire Thanks. I'm routing a save file to get some cool shit like Blue Gauntlets, 40ish Water keys, and Ass Chest. Basically, I don't know a whole lot of tricks and the route at the moment is a mess. It involves RBA Hammer to get an item on the far right that Adult can Equip Swap with so I can cheese PG before going Child 2.
Well you don't even need an AI to do routing. It's just a bunch of numbers that can easily be solved by brute force. The hardware just determines how long you have to wait. Considering that the number of permutations is on the order of n!... you'll be waiting a while, but it would eventually give you the shortest possible route for hundo.
@@LilacMonarch You can apply heuristics (or algorithms if have them) to reduce the number of necessary checks. It's a permutation of ordered timestamps, not a combination. It can be sped up by reducing the total number of permutations that need to be checked (algorithmically) and of course by having more GPUs (and/or CPUs).
He's talking about gz, which is the practice mod for OoT. It's the white text overlay that you see him bring up throughout the video. It makes speedrun practice, route timing, glitch hunting, and a ton of other things faster and easier than using a vanilla ROM.
Just noticed ZFG hit 100k can't wait to hear him say "cool" on stream.
anywayyyyy..
Truly an event of a lifetime
Lol
At least you don't get some super over the top reaction.
@@michaeltheaegis ..I wonder if ZFG will still be playing oot in another 10 years
Looking forward to a 100% SRM BAS (brainchip assisted speedrun)
Nice
I can't wait until zfg reaches 200k subs then he can start doing 200% speeduns
WhatsUpGuys%
This intro must have increased comments by at least 100% for all OoT speedrun videos
the good thing about OOT is that there are ALWAYS more optimizations and new glitches and applications of glitches to be found. so like that new route was only 30 (or 15) seconds faster, but yeah, since there are always new things to do, there's always more time to save, it'll definitely get to be a bigger difference from the old route in time. idk if there's another srm-sized glitch left in oot, but even if not, there are always some things here and there left to be discovered.
Good thing that the intro told me to expect another OoT 100% speedrun video! I almost didn't know what I was subscribed to!
4:23 Speedrunner loses a hand but instead of a bionic hand their muscles get electrically linked to the inputs of a Gamecube controller and suddenly speedrunning is their superpower. ;)
Oh shit another ZFG’s 100% speedrun video, where will the saga take us next?
imagine if people start requiring medical records to check for tas implants in order to verify the integrity of a run
And then robots become self aware and complain about discrimination
implant categories o_o
What's up guys ZFG here, bringing you another OOT 100% speedrun video
The new “show feet”
lol some idiot marked that as an intro to skip for SponsorBlock. I was wondering why I didn't hear it.
What's up guys, Dark ZFG here bringing you another OoT any% speedrun video
What's up guys, Dank ZFG here bringing you another OoT dank% speedrun video
Cursed
What's up guys, Casual ZFG here bringing you another OoT Randomizer video
A full Dijkstra algorithm on the node graph of OoT could do it, it would also take a long time to run and would be exhausting to represent OoT as a linked graph.
This is an interesting graph problem. A variant of the traveling salesman but going to certain nodes (getting an item) unlocks additional edges.
I think it’s not completely unreasonable to setup the data structure. It vaguely reminds me of an old ACM programming contest question with moving submarines. This is still way harder than that, mind you.
Would have to be greedy and wouldn’t give THE best answer but could give a pretty good one.
I don’t know enough about the OoT speed run to just make it, though.
Where do routing discussions happen? I may be biting off way more than I can chew but I’d like to take a stab at such an AI.
Rando already needs to generate a graph of what things depend on eachother and then walk it to determine if the seed is beatable. That seems like a good place to start.
To fair, AI is not that smart as people think it is. It would probably end up creating a impossible to beat route or a very bad one that actually waste more time instead of saving. Lol
oh yeah like it would be able to figure out some frame-perfect un-pause-buffered not-human-viable strats, but how the hell would an AI figure out something like text transfer glitch lol. maybe if you had an AI running OOT for two decades, the same amount of time that humans have been running it lol... but yeah
when i was a kid and i first learned about speedrunning though, that is what i thought TASes were haha. i thought they were like a computer program beating a game in the most perfect way possible, not that they were speedrunners painstakingly recording each action over and over for hundreds of hours in order to get something perfect lol, advancing like a frame at a time for many tricks. i like the reality of TASes more lol cause damn that really is something
@@gwen6622 SAAAAME LMAO
Honestly as an AI engineer, I’ve actually expressed a lot of interest in building speedrun and randomizer routes using AI and algorithms. I’d really love to work with OOT speedrunners and tech experts on this.
I guess we wouldn’t have the AI finding glitches eg srm or text transfer, but we could characterize where these glitches are possible and create dependency graphs to determine the best routes using the tricks. That’s what routing is-not figuring out tech, but applying the tech in succession after we already understand it.
Although some things might be computationally lofty (okay when we can let an optimizer run for days-we don’t need a new route every hour), a lot of the choices for a route we can tune to avoid impossible tricks or schedules unless they’re absolutely necessary. Not everything in AI is just a black box, AI’s are as smart as we build them to be, and I think computing optimal routes that are also RTA is definitely within feasibility. We would just need people to put the work in.
@@gwen6622 it would need to force it's way through the game and probably never figure out how the sandstorm after GTG works or stuff like slingshot and what stuff you can and can't shoot
@@MichaelSchwartzJazz I've always wondered where the AI parts come in since it seems like a lot of those dependencies and edge weights would need to be supplied by us, at which point it seems less like an AI problem and more like brute-forcing a travelling salesman problem.
My dude, routing is literally TSP.
the thing that makes the epic gamer intro better is the bombchu bowling music playing in the background
I started this video with 1.5x speed on somehow, and it was hilarious. Love to hear the intro moving so quickly because he sounds so over it lmao
By this point you need a PHD in OOT Routing
I mean if someone wanted to we could just run a computer program to optimize a route. It would just take some time to find the time between every possibility.
The idea would be to make a graph (like vertices with lines connecting them) and add the time to go from one point to the next as a weight. Then run a program which finds the optimal path. There may be some problems with conditionals, like if there's an action you can take with multiple possibilities but once you choose one you can do it again, but that could be sorted out fairly easily by just making a bunch of graphs with only one of those possibilities and optimizing them all, then choosing the best one of the bunch. The other difficulty is bombs, but idk there's probably a way.
Yeah, basically this would be a fancy traveling salesman problem. Edge weights based on items in inventory.
That by itself wouldn’t be too hard I don’t think, though it would be tedious to setup and debug (in my experience with such things anyways). From there one could build on top with other fun routing tools, maybe.
@@SilkNeon What do you mean the traveling salesman problem isn't hard? The entire point of that problem is how exceptionally difficult it is for computers to solve it.
@@TheLastScoot I mean, it’s not too hard to get a pretty good answer. It’s hard to get the best answer. But we use greedy algorithms to approximate an acceptable answer in these all the time.
When I say not hard I mean it more in the sense that it’s used as a setup to learn greedy algorithms in college. Meaning it could be of value in routing, I think.
@@SilkNeon Oh if it's just about getting a good enough one then fair enough, but then it becomes a question of whether that even ends up outperforming speedrunners coming up with routes or not. Sounds interesting at least.
@@SilkNeon you might be underestimating the complexity of this particular problem. as you note, finding _a_ solution is practical, well-trodden ground, even specific to OoT in some capacity wrt randomizer logic. optimizing that same problem for time is a wildly different beast.
consider that the TTYD community has crowdsourced a solution to the recipe order for their 100% TAS using a highly parallel bruteforce graph search program. there are 57 recipes that can naively be completed in any order (pre-optimization, search space order 10^76), but it's ultimately just two consecutive trips to the same NPC, and the current leading candidate is just over a minute long. check out this video if you're unfamiliar and want more specifics: ua-cam.com/video/kpn30v5Y_YQ/v-deo.html
despite the straightforward nature of the problem, the engineering effort to implement this solver was quite significant, and it ultimately faced unavoidable technical hurdles that impaired its overall effectiveness: for example, there exist more possible routes than atoms in the universe, and so there is no method for distribution of work, as a central server literally cannot retain information on the searched space indefinitely.
comparatively, the graph of OoT -- even excluding SRM -- is incomprehensibly complex, with (at least) hundreds of additional nodes and variable edge lengths owing evolving item availability (including bomb drops!), the day/night cycle, etc. we can dream of a meta-OoT that could ingest the game's source to programmatically produce such a graph/model of the game world, but each and every added degree of freedom is another step toward the (present-day) impossible solely on a computational basis, too.
it is true that we're not seeking TTYD's single-frame level of optimization, but it would still need to be reasonably precise: consider how incremental this new route wound up being relative to the time invested. I don't mean to totally kill your enthusiasm, but I intensely doubt it's feasible for OoT to have a wholly AI-generated 100% route on a multi-decadal timescale. that said, if you're into the concept, there are other speedgames out there that would for sure offer more fertile grounds for such an effort, and I would be very excited to see more of this style of approach within the speedrun community.
February 20, almost caught up. Damn, what will we do when clips are finally fully caught up
Nice another oot 100% speedrun video
Wanna catch the 100% oot speedrub video? Then you gotta SMASH that subscribe button
“I forced a bot to watch 1000 hours of oot speedruns.”
At least you have a basis for improvements of the route...
the intros are hype
"Epona's space helmet"
Three words I never thought I'd hear together, but I'm hype for it
Quick question, if I put a bottle in Nayru's Love slot before getting NL, will I be locked out of getting NL? I seem to remember something about some items not being given if you already have something in the slot.
Oooh, that's a good one. As a kid I always only Bottle Duped over Deku Nuts because that was safe and I never used them except on Ganon 😆
No, I believe it's just fire arrows that act like that.
@@magmivire
Thanks. I'm routing a save file to get some cool shit like Blue Gauntlets, 40ish Water keys, and Ass Chest. Basically, I don't know a whole lot of tricks and the route at the moment is a mess. It involves RBA Hammer to get an item on the far right that Adult can Equip Swap with so I can cheese PG before going Child 2.
11:33 - how do you get 'above' Ruto in that room? like, is there a way (without GZ or other 'cheats') that even a noob like Me can pull off? xD
I hope you're not getting burnt out on this, remember to take actual brakes zfg...
Just the title alone makes me want to make an AI for routing 😅
10:01 Why'd you have to go and make things so complicated!
That intro tho ❤️
darkredketchup
Well you don't even need an AI to do routing. It's just a bunch of numbers that can easily be solved by brute force. The hardware just determines how long you have to wait. Considering that the number of permutations is on the order of n!... you'll be waiting a while, but it would eventually give you the shortest possible route for hundo.
Let me know in 50 years how that went
@@LilacMonarch You can apply heuristics (or algorithms if have them) to reduce the number of necessary checks. It's a permutation of ordered timestamps, not a combination. It can be sped up by reducing the total number of permutations that need to be checked (algorithmically) and of course by having more GPUs (and/or CPUs).
I'll optimize the route on a quantum computer for you probably need about 20 years
Congrats on 100k
Hello
ZFG should really give dark souls 1 a try! I always thought that the game took a ton of inspiration from OoT.
"Eldin Ring is coming out soon."
Maybe ZFG would pay Dry if he'd be a little more timely with these highlights.
Wait so is the MM Lotad the one he premiered with Keeta? Damn I thought there was another one
Epic
When's buns? 🐰
sooo 13:00 This Mod is for what exactly ?
He's talking about gz, which is the practice mod for OoT. It's the white text overlay that you see him bring up throughout the video. It makes speedrun practice, route timing, glitch hunting, and a ton of other things faster and easier than using a vanilla ROM.
@@kalebroark87 yeah ^^ i know. At my time, the whole next Minute he is explaining that several Times in a row ^^
100K
oh!
Is ZFG bringing us another OOT 100% speedrun video??? cool
oot 3d pog
bruh