Rabbit CEO Jesse Lyu defended the Rabbit R1 for the last week, telling Gizmodo and other outlets that it’s “not an Android app.” Now, he may have to defend the next fatal claim for an AI device: that it’s not just a ChatGPT wrapper. The push-to-talk feature on Rabbit R1s experienced a brief outage on Thursday night, which appears to be caused by an OpenAI API outage that occurred at the same time, first spotted by The Verge.
“There is an issue with PTT queries not responding and we’re working on it urgently,” said a customer support account in Rabbit’s Discord channel on Thursday at 6:12 pm ET. “Vision mode is still operational. Team is working on it now.”
As complaints of error messages filled Rabbit’s “bug-report” Discord channel, OpenAI reported an API outage on its status page at 6:06 pm ET. The company noted “elevated error rates on fine-tuning API models and reserved capacity.” The issue was resolved for OpenAI roughly 45 minutes later, and a Rabbit customer support account updated that its push-to-talk feature should be working again around the same time.
OpenAI and Rabbit did not immediately respond to Gizmodo’s request for comment.
It’s unclear at this time whether the OpenAI outage truly caused the Rabbit outage, but it seems likely given how the R1 works. Part of the way Rabbit achieves the R1’s relatively low price point, $200, is by outsourcing some of its AI processing. In a January interview with Fast Company, Lyu noted that Rabbit uses OpenAI’s ChatGPT to understand its users’ intentions. Then, Rabbit’s proprietary large action model (LAM) is supposed to kick into gear to do stuff for you, such as play music, order food, and so on.
Emphasis on “supposed to.” As Gizmodo’s Dua Rashid points out, the Rabbit R1 isn’t great at doing stuff. When she tried to hitch a ride via the Uber app, the R1 got her pickup and drop-off location completely wrong on the first try. When using Spotify, she found the R1 would acknowledge her command to play a specific song, but then fail to actually play the song.
These hiccups were previewed by Jesse Lyu when he struggled to order McDonald’s on stage at a demo event. The R1 took roughly a minute to process his request and then displayed a random restaurant. He then blamed the DoorDash interface, not his LAM. Several minutes later, he was able to order a chicken McNugget combo meal.
Many find Rabbit’s proprietary technology difficult to use, but the R1 is still heavily reliant on other LLMs for basic functionality. OpenAI API outages are not uncommon either. In the last 90 days, the company’s status page shows roughly 10 days where partial or major outages occurred. While it’s unclear if these two events were directly related, Rabbit’s dependency on other AI models could prove to be an issue moving forward.
