Pac-Man
A classic arcade game in which the player controls a character that eats pellets in a maze while avoiding ghosts.
Timeline (last 90 days)
In experiment 1, the researchers built a multilayer perceptron with 500 inputs and 5 outputs, trained on four games—Breakout, Pong, Pac-Man, and Freeway—and the LLM, relying solely on the network architecture and raw weights (no tools, no network access), correctly identified all of the corresponding games.
In experiment 1, Astra was given only the raw weights of four game networks (Breakout, Pong, Pac-Man, Freeway) with no hint of which games they were, and correctly named all of them.
In the MIT experiment, four tiny policy networks played grid-simplified versions of Breakout, Pong, Pac-Man and Freeway respectively; each network had 17,253 weights, and Astra got all four right, with confidence between 96% and 99%.
In the game-identification experiment, Astra correctly identified all four small networks (Breakout, Pong, Pac-Man, Freeway) using only the raw weight-number text of each.
A Reddit developer encoded the weights of a 17,253-parameter neural network as a sequence of plain-text tokens and fed them into a language model, which determined that the network was used to play Pac-Man.
All Facts (5)
In experiment 1, the researchers built a multilayer perceptron with 500 inputs and 5 outputs, trained on four games—Breakout, Pong, Pac-Man, and Freeway—and the LLM, relying solely on the network architecture and raw weights (no tools, no network access), correctly identified all of the corresponding games.
50%UnverifiedIn experiment 1, Astra was given only the raw weights of four game networks (Breakout, Pong, Pac-Man, Freeway) with no hint of which games they were, and correctly named all of them.
50%UnverifiedIn the MIT experiment, four tiny policy networks played grid-simplified versions of Breakout, Pong, Pac-Man and Freeway respectively; each network had 17,253 weights, and Astra got all four right, with confidence between 96% and 99%.
50%UnverifiedIn the game-identification experiment, Astra correctly identified all four small networks (Breakout, Pong, Pac-Man, Freeway) using only the raw weight-number text of each.
50%UnverifiedA Reddit developer encoded the weights of a 17,253-parameter neural network as a sequence of plain-text tokens and fed them into a language model, which determined that the network was used to play Pac-Man.
50%