ChatGPT plays against chess AI

ChatGPT plays against chess AI

ChatGPT played against chess AI, which shocked people in less than 15 seconds.

At the beginning, ChatGPT is black, and chess AI is white. The two sides have exchanges, and ChatGPT even takes the initiative to push back the opponent’s bishop (bishop):

195569c7 3674 4696 9c0b 079437ec9827

Looks like it’s going to be a tough fight, until ChatGPT suddenly snaps and eats its own bishop!

b38e5404 c4d5 4c1f 8142 c5a887e97a08

 Strange king rook transposition operation

This is where things start to go wrong.

Whether it is the 9th black pawn appearing out of nowhere on the chessboard (there are only 8 pawns in each black and white chess):

f04b50b3 be45 4382 b1ae 30bfff404565

Or the horse that suddenly eats the bishop on the slash:

55de271b ff22 4648 b206 cffc97323b39

 The horse can only walk the word “day”

There is no reason to ignore the rules!

Such a miraculous operation attracted a large wave of netizens to watch, and some people called out, “Isn’t this the shameless operation of my 5-year-old cousin who wants to win chess?”

format,f auto

But it can also be said that ChatGPT has learned the essence of human chess in minutes – shameless …

So who won this magical AI chess battle?

Invented a “new” set of rules

Although the battle is only 1 minute and 32 seconds, ChatGPT has formulated a new set of rules.

First , it is possible to “beat cattle across mountains”. In addition to horses, even queens, rooks and bishops can easily overstep (step over a chess piece to go behind it).

0138df1c 1e90 458f aa46 19a20069c709

Second , pieces that are not on the board (eaten or do not exist) can be revived in situ, or even eat the opponent’s pieces on the spot:

0cde6b19 1247 4a10 9ab6 678ff5823bd7

 The dead queen suddenly attacked me

Some netizens ridiculed, could this be the legendary zombie tactic?

format,f auto

Third , all the tiles can be “played against common sense”, for example, the car can not only overstep the tiles, but even go diagonally…

c4565f64 d052 46ee 9d64 cae47950bd53

Spoiler for the finale, ChatGPT lost in the end, but still in the form of violating the rules – take the initiative to send the general!

7737c4f8 c13c 493b 858a d785152bb400

Some netizens ridiculed after reading it, this is probably the Calvinball chess master.

(Calvinball, a game that is not bound by rules, the only rule is “cannot use the same rules as the previous game”)

format,f auto

Some netizens expressed sympathy for ChatGPT’s opponents: How can this be predicted? (manual dog head)

format,f auto

So, what is the history of the opponents who won ChatGPT in this brawl?

The AI ​​that acts as ChatGPT’s opponent is called Stockfish , which is also an open source chess engine with a long history.

It was developed based on a neural network called NNUE , which was released in 2008. The initial structure was very simple, which was a 4-layer fully connected neural network, used in conjunction with alpha-beta search.

READ ALSO: Apple’s Reality Pro headset doesn’t need an iPhone to support ‘air typing’

After several iterations over the years, it has reached the fourth generation of NNUE, and some optimizations have been made in the architecture:

format,f auto

Now Stockfish has also iterated to version 15.1. It is said that in the battle against Stockfish 14, the number of wins reached 9 times the number of losses.

As a classic chess AI, it has often appeared in various new AI papers over the years as a performance comparison for chess.

For example, AlphaZero launched by DeepMind compared it, saying that its performance is higher than Stockfish.

Even so, Stockfish is more than enough to beat human chess masters.

“ChatGPT is more suitable for writing a chess program”

After watching the game, many people think that ChatGPT is obviously not suitable for playing chess.

Someone tried to play a game of chess with ChatGPT and explained the rules to it when it made a wrong operation. Every time ChatGPT would take the initiative to apologize, “I’m sorry, I understand”, but still insisted on making a wrong operation:

format,f auto

Probably this is the ChatGPT version of “I was wrong, but I won’t change it” (manual dog head)

However, some netizens think that it is amazing that ChatGPT can play chess.

As a language model, it is actually not good at chess games, but more suitable for writing a chess engine .

format,f auto

In fact, some netizens really did this not long ago, asking ChatGPT to write a chess program in Python.

First let it generate a set of chessboards and print them out, the effect is as follows:

format,f auto

Then, let ChatGPT formulate the method of chess piece movement according to the rules of chess, including pawns, rooks, horses, bishops, queens, and kings.

Finally, let ChatGPT generate a function that can act as a robot bot to fight against opponents.

Interested friends, you can try this engine written by ChatGPT, whether it is easy to use~

So, what do you think of ChatGPT’s chess strength? (manual dog head)