Brilliant Moves in New Game Analysis Report

wow thanks for clearing it up u guys

I have never seen the chess.com engine labeling a move as "brilliant" in the basic analysis, only in the advanced (deeper) analysis. If the engine labeled a move as brilliant because it is better than its best option, one could find brilliant moves for the basic analysis as well. In my opinion, a move is called brilliant when it requires a certain threshold depth (defined by the programmer) to be found. Apparently, the basic analysis does not have that depth so by definition no move will be called brilliant.

A few days ago I analyzed an otb game in which my oponent made a move that gave me the option to checkmate in 7 after two sacrifices. I missed it, so the engine told me I blundered. My opponent made an irrelevant move so I got a second chance. This time I saw it, so I sacrified my bishop to start the mating sequence. The engine said the move was brilliant even though it already knew it was the best option.

I analyzed the game again in basic mode, the move was classified as "best", but not "brilliant" for the reason I mention above.

And how would the engine know that?

Imagine the computer is analyzing 23.- Bxd2, which can checkmate in 432 moves. The computer can only see 431 moves further so it says 23.- Bxd2 is a neutral move. The move is made. Now the computer is analyzing 24.-Qa1 which checkmates in 431 moves. Now the computer sees it and thinks: brilliant move! I did not see it before!

This is what some people here think it is happening. What I think it is happening is something like

The computer is analyzing 23.- Bxd2, which can checkmate in 432 moves. The computer can see 1000 moves further so it can see it, but it says: wow, amazing move! it took me 432 calculations to see. Since 432 calculations is more than the threshold Mr Engineer has put (e.g. 200) I will labeled it as brilliant.

In the first case, any engine could label a move as brilliant. A potato engine could say, wow! checkmate in 2!! this must be a brilliant move because I did not see it before. In fact, I am a potato and can only see one move further. In the second case, brilliant moves can only be given by analysis that can reach the depth required to label a move as brilliant. This is why the basic potato analysis never tells a move is brilliant

Got it, you mean that the "brilliant" label might be attached later when analyzing follow-up moves. I agree that this is not very likely, because obviously the engine had to make sure that the continuation was optimal, which eventually means that it has to analyze again with sufficient depth.

The obvious implementation for such a feature would be: mark a move brilliant if it's the best move at the given search depth, but not at a (to be defined) slightly smaller depth.

brilliant is hard

let's say U use a hint, will it give U the brilliant move?

That's exactly right!

Brilliant moves cannot be suggested by the engine. It is only assessed as brilliant after being played by human player.

I want to get a brilliant move.

In Bobby Fischer's Game of the century, his amazing queen sacrifice is not considered a brilliant move. If Be6 is not brilliant, I don't know what is!

Almost right; the engine can suggest a brilliant move, but only after deepening the analysis to a certain threshold. If it's analyzing at, say, 10 depth, it might not suggest it, but at, say, 20 depth, it could.

@hikarunaku I think you should give a little more respect to others on the thread. You should have been more specific with your first & second response because it could be interpreted multiple ways. We all agree that a brilliant move is one that is only best because of a certain depth. We don't know what that depth is anyways so it seems we also don't know if the engine can go to that depth or if the human has to. And as Numquam said, it is a useless discussion, so let's at least not disrespect others

The engine would need to be able to go to that depth to know that the move is brilliant. It wouldn't normally analyze so deep, but it would on a per-move basis to check for brilliance

you have more brilliants than most of us will ever have.

Nobody has mentioned that the computer is analysing both the players games at the same time. Whether it is being analyzed before, during or after it is all the same. There is no bias, the computer/algorithm is esssentially playing it self, in a defencive verse defencive manner, its depth is probably a draw in XX number of moves, and a brilliant move drastically lowers the remaining possibilities.   It could also be learning as it analyzes, and a brilliant move is one that has not been recorded yet. This we could test =)  

I'm not quite sure what to say. To be more clear I know precisely what I'd LIKE to say, I'm just not sure how to phrase it politely. Here goes... You're making yourself look like a tool. I've not read the entire thread (and I'm not going to), so I'm not sure if there's been a resolution. Nevertheless I still feel like chiming in.

Firstly I don't understand why you attacked the first person to clearly articulate an answer. I too reread your original comment and found it less than enlightening, no offense. It didn't really have any substance to it and only really reiterated common knowledge.

The whole reason I'm on this thread is because I'm developing my own chess AI. I was hoping to find a definitive response from chess.com but as it turns out I don't need it. The gentleman you berated had it exactly right and articulated it perfectly. If you wish I can explain how their engine *probably* works but there's ten by ten thousand ways to go about it. There's no way for me to be sure. Here's the short version, if you want the long version message me.

I suspect it's got 3 levels: a neural net, an archive, and a minimax system. The minimax is fairly basic. All it does is ascribe weights to each move and a few subsequent moves (the depth). It then decides which move is best based on those weights. It goes about that by first trying to minimize the weight (blacks picks the move that's worst for white) then maximize it (whites best move)... more or less. That is fed into a neutral net alongside all the archived games weighted by the players strength, then a "checkmate win" check, and probably a frequentcy played. That would allow the neural net to upgrade the minimax autonomously.

Lastly I'd like say a brilliant movie is most certainly this: a minimax system isn't perfect. Sometimes a really good move is disregarded early because it's initially viewed as weak. It's simply an artifact of how the system works. After a move like that is forced it can drastically alter the assigned weights. I played a move like that only a few minutes ago. The weighting (the numbers that pop up in analysis) jumped from +14.99 to -65.64, an 80 point shift. I went from "you're about to lose" to "guaranteed win" in a move. So ya.

So, after brilliant move is discovered it'll be noted as best move by the engine right?

Brilliant move can only be find by player, because the definition of brilliant moves are "moves that are not discovered yet by the engine and the moves are better than the engine's best move". Is it like this?

I just played this game and both me and my opponent got a brilliant?

I think I played rather badly to be honest

https://www.chess.com/analysis/game/live/5086060570?tab=report

Would you choose either of these as brilliant moves?

Both of you two played badly, yes. I didn’t even look at the game, just the highlighted moves.... lots of reds and orange and yellow lol. 

I know!

do you think that that move deserved brilliant though

I do not