It used to be an algorithmic game for a Microsoft student competition that ran in the mid/late 2000.
The game invents a new, very simple, recursive language to move the robot (herbert) on a board, and catch all the dots while avoiding obstacles.
Amazingly this clone's executable still works today on Windows machines.
The interesting thing is that there is virtually no training data for this problem, and the rules of the game and the language are pretty clear and fit into a prompt.
The levels can be downloaded from that website and they are text based.
What I noticed last time I tried is that none of the publicly available models could solve even the most simple problem.
A reasonably decent programmer would solve the easiest problems in a very short amount of time.
It used to be an algorithmic game for a Microsoft student competition that ran in the mid/late 2000. The game invents a new, very simple, recursive language to move the robot (herbert) on a board, and catch all the dots while avoiding obstacles. Amazingly this clone's executable still works today on Windows machines.
The interesting thing is that there is virtually no training data for this problem, and the rules of the game and the language are pretty clear and fit into a prompt. The levels can be downloaded from that website and they are text based.
What I noticed last time I tried is that none of the publicly available models could solve even the most simple problem. A reasonably decent programmer would solve the easiest problems in a very short amount of time.