I took a look at the keynote. It's good, exciting work. But similar work _will_ be done by Apple and Google etc. who will bake somewhat similar functionality into their desktops, apps, OSes, home devices and mobiles in due time. Larger corporations have more inertia.
Rabbit's device is cool. It's Agent AI model is cool. But their hardware _is_ inferior to a modern phone. Why then did Rabbit made their own device? Why not a $50 app? I think it's because Apple/Google app store restrictions prevent 'superapps' that can be scripted with new code, that run a background hidden web browser (*). A custom device bypasses app store restrictions - they can do what they want.
What I do not understand is how Rabbit can sustain free usage of their 'LAM' LLM _indefinitely_ from a one-time $199 device sale. Even I'd they do make a profit on each device, that's not a lot of money for operations. Also they claim to not be selling user data. So what will fund their operational running in perpetuity? Will Rabbit instead sell its LLM trained for free by millions of customers l, to third parties? Or will they sell insights to businesses of how customers _really_ use their products? Do they want to be bought by FAANG? License patents? I would query their business model before investing.
(*) I suspect how their device works is this: for well-known apps with APIs like Uber, it users the vendors public API. For others, their LAM LLM somehow generates a script to carry out required tasks on websites via an automated headless browser session that runs on-device ( using something like phantom.js or Chromium embedded). Maybe their scripting is based off an image-recognition-centric programming framework like Sikuli (https://github.com/RaiMan/SikuliX1/). However that would break if the website changed a lot. Maybe the shared LLM helps cope with such changes.
