When the Rabbit R1 arrived earlier this year, it was an unfinished product. Engadget’s Devindra Hardawar called it “a toy that fails at everything.” Many of the features that Rabbit promised, including its signature “large action model” (LAM), were either not available at launch or did not work as promised. Now, after more than 20 software updates since the summer, Rabbit is releasing its most important yet. Starting today, every R1 user now has beta access to the training mode, a feature that allows you to train Rabbit’s AI mode to automatically perform tasks for you on any website you can visit from your computer.
Rabbit CEO and founder Jesse Lyu gave me a training demo ahead of today’s announcement. This tool is available through the company’s Rabbithole hub, and features a simple interface for creating automation. Once you’ve logged into your Rabbit account, you navigate to the website and enter your credentials if necessary to access the service you want to train the R1 to use for you. Lyu quickly realized that Rabbit does not store any username and password you enter; instead, the company saves a cookie from your training mode session for R1 to use later. In June, Rabbit had to quickly fix a security issue that could have caused serious data damage.
Once you’ve given your automation a name and written a description for it, all you have to do is perform the task you want to automate as you normally would. Rabbit’s software converts each click and interaction into commands that the R1 can then execute on its own. When Lyu called me the training routine, he trained his R1 to follow him.
Once the software has had a chance to analyze the lesson, you can repeat it automatically before testing it on your R1 to make sure it works properly. While it’s technically true you don’t need any coding knowledge to use the tutorial, approaching it from a programming perspective will likely yield better results. That’s because you can define the steps that the software will record for you as you display it automatically. It’s also useful from a troubleshooting perspective, as you can see from the video embedded above.
After you’ve tested your automation, it’s just a matter of asking your R1 to complete a test using the training method. The resulting process is not exactly the polished experience that I imagine many people have come to expect from their mobile devices. The R1 will notify you of each step of the process, and it may take some time for the device to process the query. According to Rabbit, part of that is creativity. Early testers found it helpful to have the R1 communicate its directions.
I’ll be honest, it’s hard to escape the conclusion that some of the R1 automation showed me Lyu, while creative, does not provide an efficient way to do more tasks than the applications that people already know, a point he accepted. I talked a lot during our phone call.
“There are a lot of jobs that don’t belong in the same place,” Lyu said. Until then, where he believes that the way of teaching will be the change is in the interaction that includes many platforms. Lyu gave an example of an R1 user who trained his device to order groceries. Alternatively, that person could use the R1’s camera to take photos of shopping lists created by his wife, which the device would use to order the family’s weekly groceries from their favorite stores.
Another area where R1 can provide a better experience than a dedicated app is in situations where there are competing standards, such as the current situation with smart home automation. Say you’re trying to get HomeKit and Google Home devices to work together. You won’t need to wait for the Matter Alliance to fix things. With training methods, the R1 will run the dirt for you.
“You have to think about velocity,” Lyu tells me before planning Rabbit’s final game and training routine. Currently, R1 users can freely add the community lessons they find on Rabbithole to their devices. Lyu envisions a future where users will be able to sell their automation, with Rabbit taking a cut. Additionally, while the training model is currently limited to navigation websites, Lyu suggests that it will eventually learn how to use more complex apps like Excel. At that time, Lyu argues that Rabbit will be able to deploy artificial intelligence, which understands a piece of software that was previously designed for humans.
Of course, questions remain. One big thing is that people will pay for community studies if they can easily follow automation on their own. Here, Rabbit hopes things will play out like they did on existing app stores, with more people choosing to download the apps they want rather than creating their own. “At the agent store of the future, we hope for a similar situation where any user can teach his subject if he wants, but many people will probably find lessons or agents created by other users that meet their specific needs,” the company told me. in email.
I also asked Rabbit if the company is preparing for the possibility that some platforms may block people from using the tutorial to automate tasks on their R1. In the company’s view, bot detection systems such as CAPTCHA will need to evolve to distinguish between “good agents” such as those created by Rabbit users and malicious bots.
“When a user uses LAM to perform tasks on third-party platforms, they are logging into their accounts with their credentials, and paying these companies directly for those subscriptions or services,” the company added. “We’re just providing a new platform for this transaction to happen, it’s like you can play music on your phone and your laptop… We don’t see a conflict of interest here.”
I’m not sure if things will go as smoothly as Susuro hoped, but what is clear is that the company is closer to the future that Lyu promised at the beginning of the year – even if that future still feels years away and he could be chosen by another company. For now, Rabbit hopes that R1 users will embrace the training process with enthusiasm, as that will allow the software to be developed faster.