FRITZ 20: When the Training Machine Becomes the Harshest Teacher

In a small analysis room, the screen glowed a cold blue. The clock on the wall read 1:47 a.m. A fourteen-year-old boy put his finger on the keyboard, called up the engine's first line, then turned to me: "Miss, why does the machine say this move is good, but when I played it I lost?"

I did not answer right away. I looked at the blinking line, then at the boy's scoresheet. Both were right. The machine was right that the position led to an advantage. The boy was right that he had lost. Between those two truths sits a gap that no software can ever fill — and that gap is exactly where a training engine like FRITZ 20 is meant to stand.

People say girls know nothing about tactics. So I write for them to read.

From Hamburg in 2026 to a room in Hanoi

ChessBase was founded in Hamburg in 2026 by Frederic Friedel and Matthias Wüllenweber. At first it was a small software company specialising in storing games, but out of that database grew another product line: Fritz. In 2026, at the Intel Grand Prix in London, Fritz 3 won a formal game against Garry Kasparov. In October 2026, in Bahrain, Deep Fritz drew 3-3 with Vladimir Kramnik. Four years later, in November and December 2026, in Bonn, Deep Fritz beat Kramnik 4-2.

Thirty years after Fritz 3 defeated Kasparov, the story has changed completely. The machine is no longer an opponent to be beaten. It has become a teacher. The problem is that a good teacher and a harsh teacher are two different categories, and most amateur players in Vietnam are stuck in the second without realising it.

I remember December 2026, when DeepMind published its AlphaZero paper and the whole chess community read it like a manifesto. In January 2026, Leela Chess Zero was launched to reproduce that method in an open project. Then in 2026 ChessBase introduced Fat Fritz. In 2026, Stockfish moved to the NNUE neural network architecture. The entire chess software industry changed its blood in three years. Fritz 20 sits inside that wave, but puts its emphasis somewhere else: not playing better chess, but teaching people to play better chess.

Three promises and three traps

The promotional material for FRITZ 20 uses three words: efficient, intelligent, individual. To someone who has spent years replaying games with the sound off, those are not slogans. They are three tactical problems, and each has its own trap.

Efficient. An ordinary analysis engine hands you a list of moves. A training engine hands you a diagnosis. The difference matters: if you only read the move, you are using software as a dictionary of answers. If you read the reason behind the move, you actually learn. Watching Lê Quang Liêm and Nguyễn Ngọc Trường Sơn in international events, I noticed something interesting: neither of them stops at the engine's first suggestion. They run through three or four variations, then go back and compare with what they actually played. That process takes twice as long, and it is the entire difference between someone who studies chess and someone who copies it.

Intelligent. Here, chess and football speak the same language. Belgium 2026 did not merely play football; they drew triangles on the grass. On a chessboard, that geometry is smaller, tighter, and nobody draws it for you. A knight on f3 means nothing without a pawn on e5 and a bishop on g2 forming a pincer. A good coach does not tell you the move is good. He sketches the triangle and lets you see it. FRITZ 20 is designed in that spirit: breaking positions into themes, repeating on spaced intervals, and above all managing your opening system like a living structure rather than a memorised list.

Individual. This is the hardest part and the easiest to inflate. A personalised training programme can really only do one thing: record where you go wrong, how often, and in what type of position. It does not know when you are tired, which opponent you fear, or how you lose focus after dropping a blitz game. But it knows precisely that in your last forty games you threw away an advantage in rook-and-pawn endings thirteen times. That data is not a compliment. It is a mirror.

I often choose to write about people measured with the wrong ruler. Young Vietnamese players are a textbook case. They are judged on a handful of online blitz games, while what actually decides their career sits in quiet analysis sessions at two in the morning, ungraded and unshared.

The biggest blind spot of every engine

Here is where I want to linger.

An engine evaluating a position at +0.3 means it thinks White is slightly better. It is right. But for a player rated 1,800, holding White in that position with twenty minutes left on the clock, +0.3 can be a draw or a loss depending on whether he can see the endgame structure. A 2,600-rated player may win that position nine times out of ten. Same number, same position, two entirely different fates.

That is the first blind spot: the engine has no clock, no tournament pressure, no fear when the opponent has sat motionless opposite you for twenty minutes. Elite chess is a sport of humans with limited stamina and time, yet every number the machine produces is calculated under ideal, unlimited conditions. That is why I always tell my students: do not ask the engine which move is best, ask which move is most practical within your own time budget.

The second blind spot is more dangerous: the raw calculation ability of young players is atrophying. Fifteen years ago, a school-age player had to work out variations by hand and check them by hand. Today everything has a button. I have seen children who are brilliant at screen tactics collapse completely when asked to calculate a three-move exchange sequence in their heads. That is not the software's fault. It is the fault of how we use the software. A good teacher knows when not to show the answer, and a good training engine has to be configured by the same principle.

The third blind spot is more social. Software does not discriminate by gender, not at all. Communities do. When a fourteen-year-old boy turns on his machine and shows a deep variation, people nod approvingly. When Phạm Lê Thảo Nguyên does the same thing, the first question is whether she is using the right software. I am not writing this to demand fairness. I am writing it because I once sat in exactly that chair, years ago, and I know that repeated doubt, applied often enough, bends a learning path.

Tactics never lie; people simply misread them. An engine can only answer the question you ask. If you ask for the answer, you get an answer. If you ask for the cause, you get a cause. The gap between those two questions is bigger than whether you run version nineteen or version twenty.

What matters in Vietnam

There is a reality anyone coaching chess in Vietnam knows: the biggest barrier was never the tool. A mid-range laptop, a stable connection and a free analysis engine are enough to take a young player very far. What is missing is someone sitting beside you at two in the morning to turn your question back on you, and a training structure that forces you to return to the exact place you once failed.

That is why I read the FRITZ 20 introduction with a little professional caution, but also with a little hope. Caution, because every training product drifts easily toward feature demonstration instead of changing habits. Hope, because the core idea — turning analysis from a lookup session into a tracked, repeated, diagnostic training programme — is exactly what many students at chess clubs in Hanoi and the provinces need.

When the stands are empty, you hear the studs click on the grass. That is real football. In chess, when the screen goes dark, you hear the pieces land on the board and the clock tick. That is real chess, and that is where every machine must fall silent and give up the floor.

A question left for the next game

The question is not how strong your engine is. The question is: in your last forty games, do you have the courage to reopen the exact game you lost in square thirteen of the endgame, and ask it one very specific question? A single pass can be the answer to every doubt. So can a single move. But only if you can bear to look straight at where you went wrong without clicking away to another variation.

I still keep that habit. And every time I do, the board looks a little wider.