This is what's responsible for me falling behind in my LG games this week:

“2048”:http://gabrielecirulli.github.io/2048/

But I finally got my 2048 tile this morning, and so now (I hope) I can put this very addictive game behind me.

Or you can aim for 4096 :)

I've been playing a version called Threes. I can get pretty far just trying to consolidate in a corner, but I feel there's another level of strategy I've yet to figure out.

@ypercube: Once you “win” (get a 2048 tile), it does ask if you want to keep playing. I *am* curious if a 4096 tile exists [rather: does the program generate it?], but I didn't have the patience to do it all again. For sure there is an upper limit of a 65536 tile (if you think about it for a bit, you'll see why)! Apparently the code is open source, so someone who is smarter at this than I am could check (and, probably, improve if necessary so that there is no artificial upper limit).

@pfafulous: Consolidating in a corner was the first thing I realized. After I got good at that, I realized a corollary is to avoid, at just about all costs, “trapping” a small tile surrounded by larger tiles.

You can do 4096, and another 2048 tile, but I don't recommend losing time doing it, it is quite boring…

The key is to have tiles in order 2048 1024 512 256 on first line and 16 32 64 128 on the second one, then you have plenty of space to do 16 with remaining tiles… but I don't know what the limit is.

Anyone has theoretical ideas, except for 65536?

Everyone I know has been talking about his - but it doesn't bleeping work in any of the browsers I have on my machine. (yes, that includes iceweasel and chrome with javascript enabled.)

It works fine in both Firefox and Chrome.

You are fine that'll save you a lot of time…

You either have to swipe or use the arrow keys.

Wait- FatPhil! You said javascript! LG didn't turn it into javas-cript! Are we even more vulnerable than we used to be?

@David J Bush: No, replacing JavaScript with Java-Script was a futile defense against XSS attacks. It didn't affect site security in the first place.

@Carroll, re theoretical limits: If I get to pick exactly which new tiles appear, and where, then I think it should be straightforward to reach a board with every tile, 2, 4,…,65536. Just make sure each 2 appears exactly where you need it to count up this far in binary until the board is full.

A more interesting question might be an adversarial 2-person version of this game. I am playing as normal, but you get to pick which new tiles appear (I think the choices are just 2 or 4) and where (I think the only restriction is an empty space). You make your choice each turn after I complete my previous turn. How bad of a game can you force on me? I'm not sure how much damage you can do by trying to get the 4's and 2's to alternate as much as possible, allowing very little combining of tiles.

I think the even more interesting variation of this question is if we modify the adversarial game a little, and now you have to always give me a 2 tile, but you still get to pick where it goes. Now I think it's pretty clear that early on, I can still do pretty ok, say up to 8 or 16, but how much further can I go?

Finally, the question that is closest to the original game. Given whatever the random distribution of 2's and or 4's on each turn as specified by the program, what is the probability of getting to a given tile with an optimal strategy. But since I'm not so good with probability (beyond the basics), I'm less interested in this version.

FatPhil: it works for me in iceweasel 24.4.0.

But yeah, it's not that addictive to me, so I don't think you're missing out on much. :)

Since the randomly appearing tile is either 2 or 4, I guess the upper limit is 131072 (2^17) and not 65536.

Apparently it's just a copy of http://saming.fr/p/2048/ but they made it “fancier” (which as so often happens means “broke it”). (And as mentioned before, the family tree of related games can be traced to “threes”.)

The movement/coalescing rules are unintuitive. I would have presumed that when moving left, the leftmost piece moves first, but no, it's the rightmost piece that moves first. I can't say it's a satisfying game to play at all.

@Art Duval: that game already exists –https://sphere.chronosempire.org.uk/~HEx/8402/

@ypercube: you are right! I missed that.

@z: cool. Now another game I have to avoid. :)

I tried it for a little: It seemed impossible to keep the AI from getting to 128.

The misere game is a lot quicker ;-)
I've been unable to fail to create fewer than 2 16s in all of 2 attempts

Tom Schotte (movieloverxxl) has posted a 4096 on his facebook wall.

I've also got a 4096 tile, but I couldn't reach 8192.

Very addictive… I'm very glad I got to the 2048 so I can stop playing!

http://www.theguardian.com/technology/appsblog/2014/mar/31/threes-creators-2048-game-clone

I wrote my own perfectly-playable console based 2048-alike (I've changed to rules to be more intuitive to me, but I have their rules too as a command-line option) whilst on the bus travelling from Tallinn to Riga last week. It took a couple of hours. Charging 1.50 for something as simple as that is absurd. Of course they were going to be undercut. The free market economy, motherhumper, do you speak it?

Clue to the authors of “threes” - do what iD software did - first you release the free one and get people hooked on your version, nobody will be able to undercut you, then later think about how to monetise it, and you might get lucky. Sticking a price tag on what's only a few hours work right from the start was a stupid move.

But credit to them for inventing that kind of game, plenty of people seem to like it, even if it does very little for me.

I've just added a quick auto-play AI to my perl script that does occasionally get a 1024 tile (twice in eighty runs). As it's running on my mobile phone, I don't want to flatten the battery, so I won't see if it ever gets a 2048 tile, as I suspect it won't, looking at the distribution of results.

My own variant (the tiles move in the opposite order, and therefore don't stop until all together at the target side, otherwise the same, it's more like you're tilting the board and the pieces move under gravity) is easier, and I guess the AI gets 1024 about 10-20% of the time. When I have time I'll run that on one of my home servers, and see if a 2048 appears.

@FatPhil, can you explain exactly what your rules are (when you have three 2s in raw which ones collide, where the 2s or 4s appear, with what frequency)?

As I think I get 2048 around 40% of the time it is strange your game is much harder. Does your solving algo try to get the pattern of a raw ending with 2^n,2n-1…. like:

  2   2   4  8 16
512 256 128 64 32

and keeping there raws on the corner/border?

I got 8192 at 97116 overall points. My record at now is 110952. Aiming for 16384.

BTW a weird variation with Fibonacci numbers instead of powers of 2 is here:

http://www.crazygames.com/game/2584-fibonacci

My rules:
. 2 2 2 -> . . 2 4 on moving right.
8 4 2 2 -> . . . 16 on moving right, first the twos pair, then the 4s, then the 8s.
Appearance rules for the new 2s and 4s are the same.

My AI's way too dumb to have a strategy, it only has tactics. However, it's more successful than me. When I tried to encourage it to build rows of consecutive powers of 2, it got worse.

Fibonacci sounds interesting.

On the official 2048 github page, you can find the lines:

if (this.grid.cellsAvailable()) {

var value = Math.random() < 0.9 ? 2 : 4;

which means there are nine 2s for one 4… which makes it easier.

I managed the Fibonacci, it is very good.