Wow what a move...

Yeah I like quiteirate's idea. Obviously Kg2 is not the right way to go. The bishop sacrifice gives White tactics.

kg2 its fine, its the engine recommendation,
the position has plenty of options, u just need to choose a plan and play it
but bishop e3 its more of a instructive dynamic solution much better for learning
the engine gives more than 5 points to white after king g2 but It looks more confusing like with more subvariations
all you need to know is that king g2 is a bad square if you combine it with bishop e3 so if u play bishop e3 you have to put in the h1.
My idea was just to put a pawn in the 8 file something like this,

http://en.lichess.org/MOX7q0VGFBaQ
which is the continuation on this
http://en.lichess.org/9hMIHG98AMl6
But thats only one of the many solutions cause the engine could have played other stuff and then u have to think again and so on
, the game is sharp enough to let you choose whatever you like but It does seem easy to lose the initiative if You dont pay attention.

I think I was lucky because the machine did not play Knight f6 in my game, and dont know why the machine sometimes plays one thing and sometimes another,
I dont like my game the bishop x f1 seems very slow, very passive
I think I was lucky the machine did not play a better thing like
advancing pawns or knight f6, I find lichess stockfish pretty random, so although I did not like my game I had to defend it because I won, so I dont like the idea of the bishop to f1, a rook there looks more powerful and beautiful.
But thats my play which is not good which has nothing to do with king g2 variations which is the engine recommendation, I mean of a good engine not lichess engine which looks like a drunk engine.

To clarify,
I dont like the bishop f1 as I already said but I was testing the position and I found these for example
the first game
http://en.lichess.org/9hMIHG98AMl6
the second game
http://en.lichess.org/0HYNQsEPyt3X
the third game
http://en.lichess.org/CuoC92VWMnCU
etc, there are many
so maybe its actually good that the engine plays a different line everytime, so that means that if u beat stockfish, u have beaten only one of his lines, it could play another,
so a brute force approach which is what I did not does not work because I was assuming It will only play the same move,
thats why I said the g2 variations may be more complicated, of course You dont have to do the silly move I did of bishop x f1, you can take with king (on f1) which is the more natural thing
but I am sure there are subvariations, in that king f1 variation, and not only that, stockfish plays each time a different one, in case You did not know, so that means that beating stockfish is not necessarily understanding the position but only understanding one of the multiple subvariations and if You want to understand the position You have to think about all of the many responses it has, so You have to think always with engine or without engine, winning does not necessary imply understanding the whole intricacies of the position.
So being lazy does not work, You can win a game but not understand it, u maybe win a game because the machine has played a move that is comfortable with your plan
but It could have played another move that You were not prepared