If you read what ARC-AGI has stated from day one, the tests are designed to stress frontier models in tasks that humans are uniquely good at. When the models ge

If you read what ARC-AGI has stated from day one, the tests are designed to stress frontier models in tasks that humans are uniquely good at. When the models get to 100%, the next set of tasks is deployed. When they run out of ideas for how to stress a model (i.e. no more tests), THAT is when AGI is achieved. It's a great definition.