Rabbit Demonstrates General Android Smart Agent, Still Under Development

Following the failure of the Humane AI Pin project, Rabbit continues to advance its AI technology research. The company recently published a blog post and video showcasing what it calls a "general-purpose Android intelligent agent." This agent is capable of performing a range of tasks on a tablet, similar to the features promised for the R1 device over a year ago, which were never fully realized.

The demonstration did not use the Rabbit R1 device. Instead, engineers input commands into a prompt box on a laptop, which were then translated into actions on an Android tablet. The tasks assigned to the intelligent agent included searching for videos on YouTube, finding whiskey cocktail recipes in a cocktail app, gathering required ingredients, and adding them to a shopping list in Google Keep. Additionally, the agent was tasked with downloading the puzzle game 2048 and attempting to play it, which it managed despite some slowness.

In most cases, the intelligent agent performed tasks as requested, though there were occasional anomalies. For instance, it sent a poem line by line on WhatsApp rather than sending the entire content at once. An engineer suggested that adding line breaks to the instructions might help, but they did not reattempt the task.

Rabbit's AI intelligent agent is clearly still in development. Since the launch of R1 in January 2024, its actual capabilities have fallen significantly short of the claims made by founder and CEO Jesse Lyu. The company has released several updates, such as training the AI to complete specific tasks or prompting it to redesign the user interface. According to Rabbit's blog post, the demonstrated functions represent only the "core operational loop" achieved by the Android intelligent agent. The company also pledged to share more details about its "forthcoming cross-platform multi-agent system" in the coming weeks.