François Chollet, creator of ARC-AGI, shared the news that development of the ARC-AGI-4 version is already underway, with the team planning approximately seven different variants of this benchmark.
The first of these—ARC-AGI-3—is expected to be publicly released in March. We previously provided more details about it. As for the next stage—ARC-AGI-4—it is currently being developed, and it is planned to also be related to gaming platforms and environments for training agents.
Additionally, there are plans to release ARC-AGI-5, with final versions likely being ARC-AGI-6 and 7. These may serve as the concluding stages in this series.
Chollet notes that the main idea behind ARC-AGI is to create a system capable not only of performing assigned tasks but also of passing comparative tests until only those skills or abilities remain that only humans can do—those skills or capabilities that AI has not yet mastered. He believes that the moment of achieving artificial general intelligence (AGI) will occur around 2030.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby