Quick initial response: I tried solving one of those (on my own, no AI) and got it in less than five minutes. It seems like a very underconstrained problem. In my case, I filled the cube with only Y for the outside, then alternating R, O, B, and a single G in the center. After a bit of fiddling because I misunderstood the rules (e.g. R seems to complain not just about its y-line but any straight line?) it got accepted.
(The rules were:
Y: nearest G is 2 tiles away
R: alone on its y-line
O: next to at least 2 distinct colors (I think)
B: not neighboring B
G: (can't remember, but I just had 1 and it didn't constrain))
As a human, the basic idea when solving this was "I don't wan't to remember 125 constraints, let's pick something bulky to fill most of the space", and then filling the rest with colors that don't interact with the already used ones. You could make this harder by not having such easily satisfied rules as "closest G is 2 tiles away" -- maybe rather "all the colors relative to this tile are in order XYZ". You could also remove just enough tiles to keep the result constrained, but in any case if you're generating levels randomly the rules probably won't interact in interesting ways.
---
Edit: Tried another one, timed. The link is https://www.glips.net/puzzle/f2b2720c. Solution (in one line to minimise spoilers): OBOBRYOYOGOGOYOBRYOBGOGYOYRYRYRYRYRYRYRYRYBYRYRYRYRBRBRYRYRYRYRYRYRYRYRYGYRGRYRGOGOGOYOYOYOYOYOYGOGYRBRBRGRGRGBGRGRGRGRGRBGBR
This one took me 25 minutes, so I guess the difficulty varies a lot with the rules. This one had a way more interesting dynamic for me, since there wasn't an obvious filler color, so I made a bet on a broad-strokes kind of solution that then had to get modified on various scales.
I decided from the start to just bite the bullet on R and use a checkerboard pattern for the "bulk". Initially I tried to alternate G and Y with O and Y, but that turned out to be a pain with making sure that all the Gs would match with something varying on the top and bottom side, so at some point I restarted and went for Y and R. Then the problem was making sure Ys themselves wouldn't be the majority, and I had to find ways to replace them with B and G, and I scraped by with one more R than Y.
It's interesting that your approach can fail on various scales and until you get to a result, you're not sure it will work. But I object to the notion of this being too hard for humans.
Thank you for trying these! I am not surprised to hear that they are human-solvable. I’ve looked at a few and thought “that doesn’t look so hard”. There are a few ways to filter out the easier ones during puzzle creation, but personally I don’t mind the flavor of puzzles that are randomly located a wide spectrum of difficulty. In the case of Glips, perhaps I could clarify that design constraint to be “harder, in general, than appropriate for human capabilities and preferences”.
I'm curious: When designing these puzzles, how much did you try solving them yourself, if at all? I know that isn't the purpose, but it still seems like an important way of testing.
There is a fiddliness to it -- had my approach failed, I would have probably given up, because it's annoyingly difficult (for me) to keep visually track of rules like the minimum distance from yourself, and some of the time was spent figuring out the interface (I still am not sure if the z-line is really the z-line or actually the x-line and basically tried to play such that it wouldn't matter).
But in the second attempt when the overall plan wasn't obvious immediately, it was fun to try and get a handle on a plan that *might* work without knowing all the details.
I guess that as in many other games (or activities in life in general, now that I think of it) there's a tiny part of eagle-eye strategic planning and then a big part of fiddling with details that's either frustrating or enjoyable depending on how well the plan works.
---
ETA: Just noticed a curious thing: In my solution, there are actually two Gs next to each other, and it works only because there happen to be no Rs complaining about that. This goes to show how much of the solution is just fiddling with things and responding to the errors thrown -- my solution works for a different reason than I initially thought.
Quick initial response: I tried solving one of those (on my own, no AI) and got it in less than five minutes. It seems like a very underconstrained problem. In my case, I filled the cube with only Y for the outside, then alternating R, O, B, and a single G in the center. After a bit of fiddling because I misunderstood the rules (e.g. R seems to complain not just about its y-line but any straight line?) it got accepted.
(The rules were:
Y: nearest G is 2 tiles away
R: alone on its y-line
O: next to at least 2 distinct colors (I think)
B: not neighboring B
G: (can't remember, but I just had 1 and it didn't constrain))
As a human, the basic idea when solving this was "I don't wan't to remember 125 constraints, let's pick something bulky to fill most of the space", and then filling the rest with colors that don't interact with the already used ones. You could make this harder by not having such easily satisfied rules as "closest G is 2 tiles away" -- maybe rather "all the colors relative to this tile are in order XYZ". You could also remove just enough tiles to keep the result constrained, but in any case if you're generating levels randomly the rules probably won't interact in interesting ways.
---
Edit: Tried another one, timed. The link is https://www.glips.net/puzzle/f2b2720c. Solution (in one line to minimise spoilers): OBOBRYOYOGOGOYOBRYOBGOGYOYRYRYRYRYRYRYRYRYBYRYRYRYRBRBRYRYRYRYRYRYRYRYRYGYRGRYRGOGOGOYOYOYOYOYOYGOGYRBRBRGRGRGBGRGRGRGRGRBGBR
This one took me 25 minutes, so I guess the difficulty varies a lot with the rules. This one had a way more interesting dynamic for me, since there wasn't an obvious filler color, so I made a bet on a broad-strokes kind of solution that then had to get modified on various scales.
I decided from the start to just bite the bullet on R and use a checkerboard pattern for the "bulk". Initially I tried to alternate G and Y with O and Y, but that turned out to be a pain with making sure that all the Gs would match with something varying on the top and bottom side, so at some point I restarted and went for Y and R. Then the problem was making sure Ys themselves wouldn't be the majority, and I had to find ways to replace them with B and G, and I scraped by with one more R than Y.
It's interesting that your approach can fail on various scales and until you get to a result, you're not sure it will work. But I object to the notion of this being too hard for humans.
Thank you for trying these! I am not surprised to hear that they are human-solvable. I’ve looked at a few and thought “that doesn’t look so hard”. There are a few ways to filter out the easier ones during puzzle creation, but personally I don’t mind the flavor of puzzles that are randomly located a wide spectrum of difficulty. In the case of Glips, perhaps I could clarify that design constraint to be “harder, in general, than appropriate for human capabilities and preferences”.
I'm curious: When designing these puzzles, how much did you try solving them yourself, if at all? I know that isn't the purpose, but it still seems like an important way of testing.
I never once tried. My only direct contribution was looking at them and thinking "that doesn't look hard enough" or "ok, that seems hard enough".
was it fun?
Enough to try the second time :)
There is a fiddliness to it -- had my approach failed, I would have probably given up, because it's annoyingly difficult (for me) to keep visually track of rules like the minimum distance from yourself, and some of the time was spent figuring out the interface (I still am not sure if the z-line is really the z-line or actually the x-line and basically tried to play such that it wouldn't matter).
But in the second attempt when the overall plan wasn't obvious immediately, it was fun to try and get a handle on a plan that *might* work without knowing all the details.
I guess that as in many other games (or activities in life in general, now that I think of it) there's a tiny part of eagle-eye strategic planning and then a big part of fiddling with details that's either frustrating or enjoyable depending on how well the plan works.
---
ETA: Just noticed a curious thing: In my solution, there are actually two Gs next to each other, and it works only because there happen to be no Rs complaining about that. This goes to show how much of the solution is just fiddling with things and responding to the errors thrown -- my solution works for a different reason than I initially thought.