Large Action Models vs the Rabbit R1 (and others)

The Rabbit R1 was introduced during CES 2024 and it was probably one of the promised products that helped make the the term Large Action Models (LAM) more widely known compared to mainstream Large Language Models (LLM) and, to a degree, the future iteration which is Large Agentic Models (LAM).

This got me excited since this would essentially be a physical example of what an LLM stand-alone device could be. Yes, I knew it would be buggy. And, so, I pre-ordered as soon as it became possible to do so. I'll get to the unboxing later below.

Rabbit notes what their definition of what a Large Action Model as:

Whereas, the newsletter Towards AI published their mapping comparison last March of how these currently map in their estimation:

However ....

I think some earlier example of agentic model reasoning

In a paper published on Arxiv towards the end of January 2024 talked about LLM Situational Awareness. One of the examples was a generated image and an action-flow based on what was seen, including a all to restock whatever was needed for refill.

The thing is, back in 2013 a similar function was available on some refrigerators as reported on CNET: The LG Smart Refrigerator know what you have, knows what you need. But, maybe that was probably a low hanging fruit choice example. The other examples in the paper were more interesting, such as a potentially to automatically call emergency services during life threatening situations. I doubt my smart fridge, 10-years old technology as it is, would do that. But, perhaps an SOS from a smart watch? Or in the case of vehicle accidents then the onboard computer (the old OnStar emergency services comes to mind). Since everyone has a phone, then Android location services?

My point here, being that any new use of GenAI should be showing improvements from things that already exist on many consumer devices.

Once the Rabbit R1 arrived, I let it sit on the table top for a few weeks. I gave everyone the chance to either love or pan it, as well as the developers a chance to fix the pre-release bugs they could. Today, I decided to review it. What are weekends for, right?

I found it odd that Rabbit's service link calls out to a link called VPN Proxy. Maybe its to circumvent or hedge against some network filtering? Whatever it is, the result was a wait-time to get to the log-in screen for each service.

In the case of Door Dash it wouldn't even let me in because there were "numerous logins from my IP". Well, uh, oh! Me thinks that its possible that the use of that transitional VPN could be the culprit? I'll try to give it 24-hours and if it still does the same then that more or less confirms my hunch.

(to be continued!)