The idea
Parth demonstrates Jev choosing between jumping, running and ducking in the Chrome runner game. The post describes a snapshot of the runner as input and a decision with confidence as output.
Inside the decision loop
Runner state + available actions
Choose jump, run or duck
Apply the action in the game
What to take from it
Game timing is a separate concern from model response time. Keep rendering and collision handling in the game loop, and define what happens when an answer arrives after the relevant obstacle.
The post does not fully document the snapshot adapter. This listing does not treat the demonstration as evidence of native image input, and the reported latency is specific to the creator's run.
Follow the original work
This case is an editorial interpretation of a public community demonstration. It has not been reproduced or benchmarked by this publication.
Original post by @mittalparth_