It’s only weeks until a revolutionary gadget to be released, so for the sake of history, let’s join the pieces of information we have regarding this prototype device. I’m saying prototype, because it’s the first of a kind, and even if it will fail, the idea behind it is what makes it a piper of new industrial revolution.
Disclaimer: I’ll quote Rabbit team below, so when I say “we” it means “they”. All of this information is first-hand and is not my opinion. It’s not necessarily true, but what the Rabbit declares as truth. It’s here to compare with reality later, when first reviews will be published.
The most commonly used example is DoorDash, U.S. food delivery service. That’s because Rabbit representatives are quite hungry until their devices become popular, take over the market and allow them to dictate the rules.
The Rabbit device, priced at only two hundred dollars, is a surprisingly affordable piece of modern technology. What's even more surprising is that there are no subscription fees attached to it. Rabbit has confirmed that they have no plans to introduce any subscription fees, although this could potentially change in the future. The reason for this affordability is the efficiency of the large action model and the Rabbit OS, which are less expensive to run than you might think. Once you purchase the device, it's yours to use without any additional charges.
There seems to be some confusion about the difference between a large language model and a large action model. Many people assume that it's just a marketing term or a large language model with a few extra features. However, this is not the case. A large language model, based on the Transformers architecture, is excellent at predicting the next token in a series of tokens, or the next word in a sentence. But this wouldn't have worked well with the Rabbit application.
What Rabbit does is convert your natural language into the action you want to accomplish, and then performs that action on a hidden operating system. Imagine using your phone to order food from a restaurant via DoorDash. Rabbit does all of that for you, figuring out how to navigate through the interface. It doesn't connect to an API, which makes it more efficient and less costly to run, hence the lack of a subscription fee.
Jesse Liu, in a podcast, explained the difference between a large language model and a large action model. He clarified that Rabbit is not creating a large language model, which is based on Transformer and requires a significant amount of GPU on the cloud to train and get things right. Instead, Rabbit works with the best language models and small language models, and potentially open-source language models in the future. They continuously monitor the performance of these models and can switch between them as needed. This is how Rabbit OS works. More...