Future Expansion

[QNNgg]

Thank you for the impressive progress in General Computer Control.

I want to ask are there plans to expand CRADLE's capabilities to a broader range of software and games? What criteria will be used to select new targets?

What strategies are being considered to address the scalability challenges posed by the increasing number of applications and games?

Thank you.

[WeihaoTan]

Thanks for reaching out. Yes, we indeed have plans. As for the criteria, it still heavily relies on the capability of the based model. They are still struggling with outputting the exact pixel-level position, which greatly limits the performance of mouse control. And they have difficulty in understanding the game/software-specific artifacts. So if you want to check whether Cradle can be developed into a new game/software, you need to check whether the base model can somehow (no need to be 100% correct) understand the screenshot and output precise keyboard and mouse actions. Then Cradle can greatly enhance the raw performance.

As the cutting-edge base model improves (e.g., GPT5) or the models designed for computer use appear, more and more games/software will be handled. Besides waiting/developing new models, users can also introduce prior knowledge to temporarily improve performance through prompting and tool use.