Random Blindfold Cubing Discussion

[Ollie]

It went very well! I realize that I dropped the ball on posting about it. I've had a lot of things going on in my life lately, the biggest one being that my girlfriend Katie and I are engaged! We've been busy as of late spending a lot of our time together, and we've really started wedding planning in earnest now! I'll try to get a short writeup about the cube talk as well as the powerpoint posted by sometime this weekend, I should have some time before Monday.

As for the talk, I still have the powerpoint presentation I used, and I can mention after what I remember of it too. My goal was not to have people memorize any specific commutator, or commutator class or anything like that. I did want them to walk away with the process, which I described as:

1) Identify a slice, then come up with a series of moves to destroy exactly 1 piece on that slice
2) Turn the slice
3) Undo the moves you did to destroy the piece
4) Undo the slice turn

Basically I showed them an ABA'B' commutator, leading to things like Toss Ups, Drop and Catch, etc. I then demonstrated tons of toss up and drop and catch style commutators on a computer cube projected up on a screen.

Sorry for the delay, and I will try to have at least a short description up by this weekend!

Congrats on the engagement

It sounds like a pretty good process, did your 'students' have much success applying it to an unsolved cube? I'd definitely be interested to see the PowerPoint, I'm sure others would be too

 

[Username]

Tomorrow is 4 months since I learned M2/OP, I want to celebrate it with some big attempt. Should I try a 2-5 BLD relay attempt?

 

[Mike Hughey]

From my post in the 6x6+ BLD rankings thread, megaminx is about 4x4x4 + 3x3x3, and gigaminx is about 8x8x8 + 4x4x4, or 7x7x7 + 5x5x5 + 3x3x3 + 2x2x2. (2x2x2 has about 0.334 as much information as a 3x3x3.)

I'd like to apologize. I saw your posting over there, and I guess I just sort of hoped it wasn't a true representation of information (taking into account the extra possibilities per piece). (Since megaminx is so hard for me.) I realize now you had already largely answered my question. But it's nice to have Chris's new calculations anyway, for comparison's sake.

 

[qqwref]

1) I'm thinking that the memo methods we use for bigcube BLD are largely based around 3x3x3 and/or wing cycles. This may not be the best way to approach centers because the multiple identical centers will mean you memorize more information than you need to - and as the cube gets larger more and more pieces become centers. So I'm wondering if one of those 1-6 based systems would work better. For each orbit you track cycles based on just remembering what face it goes to, and keeping track of how much of the face is solved using your fingers. So instead of memorizing something like UQBK... you would memorize something like 462153... There are now roughly two digits for each cycle (plus some more to deal with solved pieces), which is IMO an improvement over two letters. You can memorize two or three digits with one image, or maybe even 4 if you're crazy. I feel like for very large cubes this might pay off because of the slower algs (so a little overhead in figuring out what pieces are being cycled doesn't matter as much). It should decrease the amount to memo. What do people think - is this a good idea, has it been thought of before, etc.?


2) Maskow's best sub-hour blindfold is 35/37 (I think). So he solved 35 cubes in an hour. That is 700 pieces, which is just a little less than a 12x12x12 cube has. It is also about the same amount of information as a 14x14x14 cube (which is actually just a hair over 36 cubes' worth). Just thought that was kinda interesting - in terms of amount of memory or number of pieces memorized and solved, multiBLD is far far more advanced than bigBLD.

 

[A Leman]

1) I'm thinking that the memo methods we use for bigcube BLD are largely based around 3x3x3 and/or wing cycles. This may not be the best way to approach centers because the multiple identical centers will mean you memorize more information than you need to - and as the cube gets larger more and more pieces become centers. So I'm wondering if one of those 1-6 based systems would work better. For each orbit you track cycles based on just remembering what face it goes to, and keeping track of how much of the face is solved using your fingers. So instead of memorizing something like UQBK... you would memorize something like 462153... There are now roughly two digits for each cycle (plus some more to deal with solved pieces), which is IMO an improvement over two letters. You can memorize two or three digits with one image, or maybe even 4 if you're crazy. I feel like for very large cubes this might pay off because of the slower algs (so a little overhead in figuring out what pieces are being cycled doesn't matter as much). It should decrease the amount to memo. What do people think - is this a good idea, has it been thought of before, etc.?


2) Maskow's best sub-hour blindfold is 35/37 (I think). So he solved 35 cubes in an hour. That is 700 pieces, which is just a little less than a 12x12x12 cube has. It is also about the same amount of information as a 14x14x14 cube (which is actually just a hair over 36 cubes' worth). Just thought that was kinda interesting - in terms of amount of memory or number of pieces memorized and solved, multiBLD is far far more advanced than bigBLD.

1) I think this is a bad idea for a couple reasons.
A big benefit of letter pairs is that they translate to commutators very nicely and there is almost no thinking. It also describes a specific target. If someone used a 3digit system, then this would restrict the recall because you would need to remember both images at once, spit it in the middle and then do the cycle. 6^3=216 which means more repeat images. you would need to keep track of which targets you already shot to(this seems error prone) on a face and isolated 2cycles would be harder to check for.
2) Maskow is somewhat in his own league for Multi, but he has not done BigBLD seriously(that I know of). Anyway, don't you agree that there is a lot more pressure for accuracy with BigBLD,there is more to keep track of for each subset(~24 vs ~12), Big cubes are more difficult to turn and really big cubes are too expensive to invest in(especially since they are not an offical event). I feel like those will be holding things back. I'm done ranting and I'm sorry if I sound pessimistic. You did bring up some great questions.

 

[antoineccantin]

1) I think this is a bad idea for a couple reasons.
2) Maskow is somewhat in his own league for Multi, but he has not done BigBLD seriously(that I know of). Anyway, don't you agree that there is a lot more pressure for accuracy with BigBLD,there is more to keep track of for each subset(~24 vs ~12), Big cubes are more difficult to turn and really big cubes are too expensive to invest in(especially since they are not an offical event). I feel like those will be holding things back. I'm done ranting and I'm sorry if I sound pessimistic. You did bring up some great questions.

I'm pretty sure he vowed never to try bigBLD. I'm not sure why.

 

[DrKorbin]

1) .....

Generally this idea is not good because if you use letter scheme with 24 letters, you can memorize an alg for each word for x- and t- centers. These algs almost always can be used for obliques.
For example, you know that Ubr>Fdl>Dbr can be executed as U [r2, U l' U'] U'. A similar alg can be used for corresponding t-centers: Ub>Fd>Dr - U [r2, U M' U'] U'.
So you know that when you encounter Ubbr>Fddl>Drrb in obliques, you can use U [r2, U 3l' U'] U'. And all these three 3-cycles are memorized with one and the same word (in my letter scheme at least). You just learn that when you got this word in x-, t- or oblique centers, you execute one of several similar algorithms.
So what I'm trying to say is that if you use "standard" letter scheme for centers in 4x4 and 5x5 then you should also use it for all centers in very big cubes. And I would recommend the "standard" letter scheme for 4x4 and 5x5 because you can memorize alg for each word and become fast

has it been thought of before, etc.?

For example, Clément Gallet uses visual memorization and he memorize just the face of the center. If the center is solved, he memorizes the face several times.

 

[cmhardw]

All this discussion had got me thinking about how well of a comparison this Scramble Space method is that we've been using. Essentially what we are doing is taking a logarithm base 43252003274489856000 to see how many 3x3x3s a certain puzzle is equivalent to. The great thing about this approach is that for 3x3x3 MultiBLD this method gives you the exact number of cubes attempted.

Taking the same idea and applying it to a large 3x3x3 multi can also, in theory, give a comparison of a multiBLD attempt using different puzzles. Take Maskow's recent 39/41 multi attempt (Amazing result!). I tried to find a nearly equivalent multi using other puzzles using the Scramble Space method (logarithms) and I found one that is very close!

A multiBLD attempt consisting of three gigaminxes, two 2x2x2s and a final 2x2x2 where you only memorize the orientations of the last layer corners is very closely the equivalent Informations Space of 41 3x3x3s.

To be clear about my proposed multi, for the third 2x2x2 all pieces would be in their solved positions, a specific last layer color would be decided upon prior to the start of the event, and the corners with that decided upon last layer color could be oriented in any of their 3^3 possible orientations.

The correlation for these two multi attemps is quite close:

\( \frac{41*log(43252003274489856000)}{3*log(\frac{20!*30!*(60!)^3*3^{19}*2^{26}}{(5!)^{12}})+2*log(7!*3^6)+log(3^3)}\approx0.9998 \)

To those who have done gigaminx BLD, does this seem reasonable? Is the perceived difficulty of doing a Gigaminx vs. doing a 41 cube 3x3x3 multi the same? According to the possible information space they are nearly equivalent, but is the difficulty in actually trying 41 3x3x3s the same as trying 3 gigaminxes plus the 2x2x2s to even out the information space?

 

[A Leman]

To those who have done gigaminx BLD, does this seem reasonable?

Has anyone even done this? I think that Mike is the only person practicing Gigaminx and he has not had a success yet. All that I can say is that for MegaBLD, I needed to memorize one sticker at a time because I don't have a 60^2 image list(any have no plans of ever making one). In terms of actual difficulty, execution would be awful. Both attempts sound really difficult!

 

[Mike Hughey]

Regarding multi vs. gigaminx:
---------------------------
I think I am the only one crazy enough to try a gigaminx yet. The minxes are just so different from everything else. And most importantly, I have done absolutely nothing to optimize my memo method for gigaminx, so it is much harder for me than it theoretically should be. And I'm not willing to put in the effort to revamp my memo methods just to do minxes.

I really doubt I'll ever do a gigaminx in under 3 hours (my best attempt yet is 4 1/2 hours). But I could certainly do 41 cubes in under 3 hours with some work - I almost could do them now. So that's at least a 3:1 difference in difficulty. But it just doesn't match because minxes are so much harder, at least for me.

I suspect that minxes are harder for everyone anyway; it's much harder to maintain the proper orientation on a minx. I use a specific pinch grip to try to maintain it; I sometimes move away from the grip for short periods of time, but I always go back to it, to try to maintain orientation. But there's the constant fear you're going to misorient. And note you couldn't use the method qqwref mentioned for centers on a minx - not enough fingers (and it would greatly increase the likelihood of misorienting).

Regarding qqwref's idea about center memo:
------------------------------------------
There has been some discussion about this method, and several people who have worked on it. I've played with it just a little, but I don't think I could manage to keep track of the pieces with my fingers during the solve well enough. That problem is the main reason why I haven't tried it.

Note that this method has another possible sub-method, which is to memorize each face directly instead of memorizing the cycles. You'd simply memorize a single image telling which face each piece in that particular orbit on that particular face needs to go to. It probably wouldn't be that hard to decode on the fly, since there's really not much information now, and it would be incredibly fast to memorize (one of the good people could memorize a face for a given piece type in less than a second, probably - so 6 seconds per piece type!). This is actually the main reason why I wanted to try this approach. But I just haven't worked out how to keep track of where I am on each face during the solve - I just don't seem to be able to figure out how to make the fingers work, and I can't think of another good way to keep track.

 

[bobthegiraffemonkey]

Regarding multi vs. gigaminx:
---------------------------
I suspect that minxes are harder for everyone anyway; it's much harder to maintain the proper orientation on a minx. I use a specific pinch grip to try to maintain it; I sometimes move away from the grip for short periods of time, but I always go back to it, to try to maintain orientation. But there's the constant fear you're going to misorient.

While this is a valid point, it seems like Chris' post regards information as opposed to solving difficulty. 3x3 multi execution only involves 3x3s, clearly far easier than a gigaminx for anyone. However, due to the problems most have of not having their letter pair scheme cover all the different targets the actual amount of memorised information for gigaminx BLD can be effectively higher than it needs to be.

I feel that some effort to include how optimised certain memo methods are for different attempts, as this will give a better idea of relative difficulty. Is there an easy way to take account of this in the calculations, since they are based on the number of positions and not how much information is typically memorised?

 

[qqwref]

Note that this method has another possible sub-method, which is to memorize each face directly instead of memorizing the cycles. You'd simply memorize a single image telling which face each piece in that particular orbit on that particular face needs to go to. It probably wouldn't be that hard to decode on the fly, since there's really not much information now, and it would be incredibly fast to memorize (one of the good people could memorize a face for a given piece type in less than a second, probably - so 6 seconds per piece type!). This is actually the main reason why I wanted to try this approach. But I just haven't worked out how to keep track of where I am on each face during the solve - I just don't seem to be able to figure out how to make the fingers work, and I can't think of another good way to keep track.

Wow, what a concept. Execution sounds really tough, but if anyone ever wants to just memo a 4x4x4 or 5x5x5 as fast as possible, this might be the best way to do it.

I feel that some effort to include how optimised certain memo methods are for different attempts, as this will give a better idea of relative difficulty. Is there an easy way to take account of this in the calculations, since they are based on the number of positions and not how much information is typically memorised?

As a first approximation, we could figure out how much information there is per image, and then express puzzles' information content in terms of images instead. For 3x3x3s I assume the best people can put two 1-24 numbers in each image, so a 3x3x3 works out to log(4.3252 * 10^19)/log(24^2) = 7.113 images, roughly. On a Megaminx, however, it's probably fair to say the best people can put one 1-60 number in each image, meaning that it works out to log(1.0067 * 10^68)/log(60) = 38.244 images. That is about 5.377 times as much as the 3x3x3. (Note that these numbes don't correspond to the number of images memorized, but rather, the information content of the puzzle divided by the information content of each image. Real memo techniques should take more images than this, on average. Hopefully not much more, though.)

 

[mycube]

what is the biggest amount of cubes someone did in a multibld-attempt with visual memo?

 

[Bhargav777]

what is the biggest amount of cubes someone did in a multibld-attempt with visual memo?

4 by Chester I guess

 

[mycube]

4 by Chester I guess

I know at least 2 people who did more (jan bentlage 9 and me 8)

 

[Ollie]

After speaking to Joey this weekend I decided to try some visual solves. Only visual memo is pretty damn hard nowadays...

2:16.44 single and 6:14.74 1/2. I had to really try hard to not say the letter pair images in my head as I memorized the targets. Tried to use mostly lines, tapping and dots to no avail. May try this again later

 

[DrKorbin]

2:16.44 single and 6:14.74 1/2

Lol I guess this is not 4x4, right?
Why did you try that? You think visual memo has more potential than letter pairs/stories/journey/audio?

 

[Ollie]

Why did you try that? You think visual memo has more potential than letter pairs/stories/journey/audio?

Haha, noo It certainly has value when combined with letters/audio/journey method etc (especially for big cube centers.) I guess it was mostly for fun, and also to try and understand how you could memorize without letters, and it's pretty plucking hard!

 

[mycube]

and it's pretty plucking hard!

well, no for me it feels normal and hard to image to memorize with something else. Depends on the way you learned it

 

[Bhargav777]

How do you guys memorise for any letter pair starting with Q? QH, QV QA, QE . TOO PAINFUL!