"SIMA isn't trained to win a game; it's trained to run it and do what it's told," said Tim Harley, Google DeepMind researcher.
In a blog post, Google noted SIMA's ability to play games without requiring custom APIs or source codes access.
Collection
[
|
...
]