• Welcome to the Speedsolving.com, home of the web's largest puzzle community!
    You are currently viewing our forum as a guest which gives you limited access to join discussions and access our other features.

    Registration is fast, simple and absolutely free so please, join our community of 40,000+ people from around the world today!

    If you are already a member, simply login to hide this message and begin participating in the community!

Will Advanced Blind Methods Become the Best Speedsolving Method for Humans?

OreKehStrah

Member
Joined
May 24, 2019
Messages
1,435
YouTube
Visit Channel
Some people already know that I'm super interested in methodology and what techniques and methods will become useful or deemed best in the future, so I'm always thinking about what might be best.

I currently believe that the best method for speedsolving for humans will be advanced blindsolving methods like 3-style, and maybe in the future, 5-style.

There is already a benefit to being able to use a normal method and 3-style for solves that are better for each method. Having a scramble with a bunch of random solved pieces is trash for CFOP, Roux, ZZ, etc, while they are pretty nice for 3-style. This is the first reason I think blindsolving is going to be a very powerful method for speedsolving.

Now for the stronger part of this argument. If you look at the robots that can solve the cube in a second or so, they show that the cube can be physically solved in incredible speeds, although I will concede that robots don't face the ergonomics vs movecount issue humans face. However, if you look at how these robots work, they offload the solving part of the cube to the beginning and find the entire solution, and then execute the solution. It is universally agreed that any solve that is completely pauseless, or 1-look, is going to be faster than a non-1-look solve. The only practical way we currently have to do this would be to use a blindfolded method and memorize the entire cube and plan the solution in the inspection time, and then execute the solve in a 1-look, pauseless manner. Therefore I believe that in the future, using a blindsolving method will be the best method a human could use based on what techniques are available and known currently.

I'd love to hear what other members of the community think about this, and especially those that already use 3-style.
 

xyzzy

Member
Joined
Dec 24, 2015
Messages
2,876
However, if you look at how these robots work, they offload the solving part of the cube to the beginning and find the entire solution, and then execute the solution.
That's because that's where the optimal split between solving time (as in, getting a solution) and execution time is for 3×3×3 with Kociemba 2-phase. The robots turn fast, but they don't turn infinitely fast. Each move still takes a few centiseconds to execute, and spending one centisecond extra on getting a solution shorter by 1-2 moves is completely worth it; however, when trying out multiple solutions with the Kociemba algorithm, the different solutions generally don't share the first few moves, so you can't execute a move and simultaneously search for better continuations. Ergo, compute the whole solution up front and execute that.

If it were to be paired with a fast optimal solver (e.g. something that uses 20+ gigabytes of precomputed tables), then I'd expect that the best split would be something like applying 3-5 moves of a good 2-phase solution, and optimally solving the rest while those 3-5 moves are being physically executed. This wouldn't be "1-look" anymore. (It also probably saves like half a move on average, so it's hard to see people caring about this when they can just make their robots' motors faster instead.)

(This is under the assumption that no separate inspection time is granted. If there is inspection time, then 15 seconds is plenty of time to get an optimal/optimal+1 solution.)

It is universally agreed that any solve that is completely pauseless, or 1-look, is going to be faster than a non-1-look solve. The only practical way we currently have to do this would be to use a blindfolded method and memorize the entire cube and plan the solution in the inspection time, and then execute the solve in a 1-look, pauseless manner.
All else equal, yes, 1-look is better than not-1-look. However, 3-style has a much higher average move count than Roux/ZB/ZZ. You'd need a higher turning speed to compensate for that, and there's a physical limit to how fast one's hands can move. In contrast, lookahead/recognition is something that is 100% mental and (theoretically) can be trained to match higher turning speeds.

But that said, 3-style algs can also be executed faster than usual speedsolves because they're fingertricky comms, so it's hard to say and maybe I'm completely off track here.
 

porkynator

Member
Joined
Oct 27, 2010
Messages
1,322
Location
Belluno, Italy
WCA
2011TRON02
YouTube
Visit Channel
I think the difference in movecount is too large. 3-style uses about twice as many moves as a speedsolving method such as CFOP. Having the full solve planned out in inspection is a nice advantage, but this does not let you turn twice as fast. If you look at the fastest speedsolvers you can clearly see that it is possible to think while turning very fast :) Moreover, some algorithmic steps (OLL, PLL, CMLL) require very little recognition time and can be solved at the same speed that one does with 3-style.

I think future speedsolving methods (including "future versions" of CFOP, Roux and ZZ) will rely more on tricks that improve efficiency while not sacrificing ease of recognition and ergonomics.

So my answer is no, unless someone comes up with a very clever blindsolving method that uses ~25% less moves than 3-style and has the same ergnomics.
 

GAN 356 X

Member
Joined
Aug 10, 2019
Messages
1,396
Location
Somewhere in the cubing universe
I have thought similar things before. If they perhaps optimised the blind methods it might be possible to lower the move count, make the algorithms more ergonomic, and so on. Keep in mind I know little to nothing about blind solving. :p
 

Billabob

Member
Joined
Jul 12, 2018
Messages
180
3-style is simply too move-heavy, as others have already said. Here's Jack Cai, one of the best BLD solvers, getting an 11.95 average with 3-style:

The only feasible way to learn 5-style is to find a method to create algorithms on the spot. You can't learn that many algorithms.
 

OreKehStrah

Member
Joined
May 24, 2019
Messages
1,435
YouTube
Visit Channel
3-style is simply too move-heavy, as others have already said. Here's Jack Cai, one of the best BLD solvers, getting an 11.95 average with 3-style:

The only feasible way to learn 5-style is to find a method to create algorithms on the spot. You can't learn that many algorithms.
To be fair, people said ZBLL was too much to learn and then people started learning the whole dang thing. So it would be insanely difficult but it think it would be reasonable to learn basic cycles using 5-style. This is also not in the perspective of the immediate future. Im interested in what will be fast in the next few decades, if not a century lol

I think the difference in movecount is too large. 3-style uses about twice as many moves as a speedsolving method such as CFOP. Having the full solve planned out in inspection is a nice advantage, but this does not let you turn twice as fast. If you look at the fastest speedsolvers you can clearly see that it is possible to think while turning very fast :) Moreover, some algorithmic steps (OLL, PLL, CMLL) require very little recognition time and can be solved at the same speed that one does with 3-style.

I think future speedsolving methods (including "future versions" of CFOP, Roux and ZZ) will rely more on tricks that improve efficiency while not sacrificing ease of recognition and ergonomics.

So my answer is no, unless someone comes up with a very clever blindsolving method that uses ~25% less moves than 3-style and has the same ergnomics.
I kinda figured that the move count would cause problems. I’m by no means an expert at blind so I wanted to get community feedback.
 
Joined
Oct 7, 2018
Messages
335
WCA
2017OWEN01
the other thing that no one is taking into account is thr moveset:
CFOP: predominantly R U F L gen
ZZ predominantly R U L GEN
ROUX: predominantly R U RW M GEN
3-style: literally every move including B S and E moves
you can't get the same tps on E' [U', R' E R] as a tperm
 

efattah

Member
Joined
Feb 14, 2016
Messages
711
If I'm not mistaken, a few people in the community have published verified videos where they 1-looked the entire solve with CFOP. Sure, it takes them many minutes of inspection. But then you could have an autistic savant who might be able to do it in 15 seconds. Which means it would be possible (for a few rare people in the world), to 1-look using Roux or other top methods. In which case the argument falls apart because you have both 1-look and low movecount, vs. blindsolving methods which offer 1-look and high movecount.
 

OreKehStrah

Member
Joined
May 24, 2019
Messages
1,435
YouTube
Visit Channel
If I'm not mistaken, a few people in the community have published verified videos where they 1-looked the entire solve with CFOP. Sure, it takes them many minutes of inspection. But then you could have an autistic savant who might be able to do it in 15 seconds. Which means it would be possible (for a few rare people in the world), to 1-look using Roux or other top methods. In which case the argument falls apart because you have both 1-look and low movecount, vs. blindsolving methods which offer 1-look and high movecount.
While that is true, that is not the norm. You mention yourself that it would be possible "for a few rare people," therefore they are exceptions to the norm. And if you know anything about statistics you should know that exceptions to a rule don't negate the rule ;)
 

parkertrager

Member
Joined
May 28, 2019
Messages
6
no offence but that is a terrible idea. 3 style although is pauseless and you'll have everything planned out allowing you to cancel and spam tps to a greater extent than zz cfop and roux. But the move count is significantly worse you would have to spam tps more than humanly possible to achieve wr potential times. 5 style may be slightly more justifiable. but id say not 5 style id assume has a much lower move count. but 5 style is a lot .less intuitive with 160000 comns. this also based off how many pieces in a cube it prob has more moves than the current main methods. also with other methods you can get a lot luckier so no bld methods are not the future.
 

sqAree

Member
Joined
Jun 10, 2015
Messages
858
Location
Berlin
WCA
2015JAEH01
YouTube
Visit Channel
the other thing that no one is taking into account is thr moveset:
CFOP: predominantly R U F L gen
ZZ predominantly R U L GEN
ROUX: predominantly R U RW M GEN
3-style: literally every move including B S and E moves
you can't get the same tps on E' [U', R' E R] as a tperm

Many 3-style algs can be really fast though. I just randomly timed [R' E R, U'] and got 0.55 after like two attempts. This would be 14.5 TPS. My T-perm that I trained for years takes me one second (maybe I can get faster singles sometimes, but not consistently). This would be 14 TPS.
Of course there are also some bad cases where getting the same TPS is not possible (same with CFOP though, T-perm is arguably one of the best cases). And corner comms are mostly pure RUD which is one of the best, if not the best, moveset.

But that aside, we'd have to have twice as high TPS on 3-style as with CFOP to compensate the movecount, so yeah, 3-style just won't get you world record times, I admit.
 

efattah

Member
Joined
Feb 14, 2016
Messages
711
Perhaps a more interesting question would be the future addition of a 3x3 category with 2-minute inspection time. That would change things dramatically. Suddenly stuff like method neutral might even be worth it, or 1-looking the entire solve. Or even more interesting, blind-solve with 4-minute inspection. i.e. you are expected to use CFOP but blindfolded.
 

abunickabhi

Member
Joined
Jan 9, 2014
Messages
6,682
Location
Yo
WCA
2013GHOD01
YouTube
Visit Channel
Perhaps a more interesting question would be the future addition of a 3x3 category with 2-minute inspection time. That would change things dramatically. Suddenly stuff like method neutral might even be worth it, or 1-looking the entire solve. Or even more interesting, blind-solve with 4-minute inspection. i.e. you are expected to use CFOP but blindfolded.

Sorry for the bump, but I agree with Fattah, that the dynamics of solving will be changed if inspection limit is changed.

I even gave a talk on Method Neutrality in Indian Nationals 2019,


I have been using CFOP+Roux for my 3x3 speedsolves since 2018, and I feel 15 seconds is enough to decide on which method to choose.

Also, I do believe that 5-style edges + RUD UFR corner comms will be the method people will be using in the next decade for speedsolving. Its an advanced blind method, but it can give execution time, almost equal to CFOP, if trained properly, and enough people work on the fingertricks of the new UF5 set.

More details about 5-style here,

BTW, Fattah, big fan of your LMCF method since 2016. Sadly, I only tried the waterman variants for a bit, and did not main the method, as my focus is on blindsolving.
 

abunickabhi

Member
Joined
Jan 9, 2014
Messages
6,682
Location
Yo
WCA
2013GHOD01
YouTube
Visit Channel
3-style is simply too move-heavy, as others have already said. Here's Jack Cai, one of the best BLD solvers, getting an 11.95 average with 3-style:

The only feasible way to learn 5-style is to find a method to create algorithms on the spot. You can't learn that many algorithms.
creating algs on the spot is not good and fast. Its always good to prepare 5-style set, and use it when we encounter a familiar case, in 3x3 or 3BLD.


There is a big difference between 493 algs and significantly more than 100,000 algs
He meant a subset of the 5-style edges algorithm. A smaller subset of of 9 mover or less is made, and its below 10k algs, and quite learnable, especially using Yo notation. It is not that good to learn the normal way, as there are no common triggers that you can use the muscle memory for.

no offence but that is a terrible idea. 3 style although is pauseless and you'll have everything planned out allowing you to cancel and spam tps to a greater extent than zz cfop and roux. But the move count is significantly worse you would have to spam tps more than humanly possible to achieve wr potential times. 5 style may be slightly more justifiable. but id say not 5 style id assume has a much lower move count. but 5 style is a lot .less intuitive with 160000 comns. this also based off how many pieces in a cube it prob has more moves than the current main methods. also with other methods you can get a lot luckier so no bld methods are not the future.
5-style has 126,720 edge cases, but not all of them are good. It is more like 10-20k cases that will give you a speedup over 3-style. Also if you consider mirrors and inverses, the alg count becomes much lesser.

And yes you are correct, the average solution using 5-cycles is similar to CFOP length solution (50-60 moves), and if you measure in STM.
 

eyeoh

Member
Joined
May 11, 2020
Messages
38
Location
Australia
I currently believe that the best method for speedsolving for humans will be advanced blindsolving methods like 3-style, and maybe in the future, 5-style.
After the original question: "Will Advanced Blind Methods Become the Best Speedsolving Method for Humans?".

I assume you meant for 3x3x3 average of 5 in competition (or ao100/1000 at home, etc), and that "humans" means either most people, or the top 3x3x3 solvers; rather than those who "quit" CFOP to focus only on 3-Style, in which case, they could end up faster with 3-Style method than anything else.

For example, Jack Cai says he's faster at 3-Style method, but I doubt he practises CFOP at all these days so I don't think the "CFOP vs 3-Style" comparison here is quite fair since they're different tools for different jobs. Sure, you could use an old hard drive to hammer in a nail, but on average, a hammer will get the job done better. Similarly, you could use a hammer to record data or basic information in a pinch, but again, the tool that's designed for recording data will be better at it on average.

As for 5-Style, I don't think relatively normal people would care to embark on this journey when CFOP/Roux is almost a guarantee (for many dedicated speedsolvers anyway) to sub-10 given sufficient high-quality training and practice. Is there a course for full 5-Style that people can complete in a finite time while not neglecting other responsibilities or hobbies? No. So until there's evidence of it actually working in practice, maybe it's better to lump it in the "beyond advanced" category, and assume "advanced" means 3-Style and nothing more.

Anyway, there's a simpler way to look at this: Observe the current best 3blders - Tommy, Elliott etc. (apologies for omissions), who of late have gotten really good really quickly. Time how long their exec takes. Is it *consistently* 5, 6 or even 7 seconds? Why not? It's not like they pause all that much amidst all that high tps turning.

It's the move count.
 

IsThatA4x4

Member
Joined
Jul 18, 2021
Messages
915
Location
UK
WCA
2022RITC01
I think, while methods like 3 (and eventually 5)-style will improve massively in the future, so will methods like CFOP, Roux, and ZZ. The standards are only going to get higher. There will be some people who learn full 1LLL and optimise it, and will be able to 1-look F2L almost every solve in the future. Recent records have shown just how much potential is still left to be unlocked.
 
Top