Chess in 72 Hours

Deep Learning Machine Teaches Itself Chess in 72 Hours, Plays at International Master Level - очень похоже на рекламу подтяжек. снова бред про brute force. к счастью среди комментирующих есть вменяемые люди.

jbrook: There are multiple statements in this article that are incorrect on so many levels that I don't even know where to begin to point out the mistakes.

The framework of the engine proposed in the paper is nothing different from a typical chess engine. It still uses the standard algorithm of minimax search to look through possible moves and prune the search trees appropriately with again, the standard technique known as alpha-beta pruning. The pruning process is guided by a function that evaluates the board positions. Improving evaluation functions have been a research area since the 50's. What's novel in the paper is that the evaluation function is trained using deep reinforcement learning -- a recently developed approach by DeepMind.

That being said, the idea of using reinforcement learning or even deep learning to evaluate board position is not new, and you can find journal papers on the topics and alike as early as the 70s if you just run a Google scholar search. So the subtitle -- In a world first, an artificial intelligence machine plays chess by evaluating the board rather than using brute force to work out every possible move -- is simply factually incorrect and does not make any sense.
...
I wouldn't be surprised to read a confused article like this that hypes a non-peer reviewed publication on a newspaper website, but how did this get onto MIT Tech review?

[выделено мной - вс]

от себя добавлю - статья проплачена. тем или иным способом. ну ланч может он ему купил, или пиво хотя бы - не знаю.

см. также у mi3ch