I noticed that the Activity button on the iPhone can be programmed to start a voice chat with ChatGPT. It is a bit slow to connect, but it is actually pretty similar to the r1 interaction.
It is probably hard to compete on raw LLM power, and the iPhone camera is so much stronger. I think that for me to prefer the r1 it will need to be doing things for me, like that spreadsheet example in the original ad demo.
A thing I’d like help with is monitoring Slack and letting me know where my attention is needed. Or Github issues. It could start a conversation with me on how to reply and then also be my interface for actually replying.
I’m sure there are tons of use cases for active assistance. Giving us API:s to the device, and the rabbit hole, as well as webhooks (I think I saw that webhooks are coming) will let us quickly explore.
I agree. I’ve linked my active button to ChatGPT 4o with a single tap of the button. Works faster and a lot more accurately then the R1 itself. Plus it can hold a conversation and not just say “I can’t do that”.
I deff agree that they should let devs explore more if its going to take them longer to build their LAM stuff. I am waiting for webhooks but also hope there could be some kind of 2 way communication with R1. I wish they could at least give some details so I could start planning some dev around it.
I think there are some misconceptions about what ChatGPT is and what the r1 is intended to be. ChatGPT is a question answering bot that also has voice capabilities. We will eventually get the conversational side of demo. (Maybe the paid subscription already has?) The r1 is intended to be an assistant that can take real world actions. They are two completely different forks of development and they serve different purposes.
That said, we don’t have LAM yet either. We’re only seeing the tip of the iceberg. Also consider the limitations that Rabbit is operating under. They have a fraction of the compute and a fraction of the dev team working on the device. However, I also know how bad ChatGPTs dev team is at building functionality into ChatGPT. Their RAG is attrocious. Their code execution was riddled with vulnerabilities. Their GPTs leak. Of course the rabbit doesn’t seem to really have any of these features. Memory is just a simple, short chat history, and they’re probably operating under tiny context windows.
I think we need to give it 6 months before we pass any judgement.Honestly, I doubt either platform is going to nail memory. I still think the r1 has an opportunity to shine with the action model, depending on how it is implemented. Any model can do function calling with the right prompts.
I used Siri Shortcuts to create a menu that opens up when I press the action button on my iPhone, and that menu includes several other shortcuts (control lights, hear a song, see passwords, access notes, etc.). One of those was a voice chat with ChatGPT. I used it a few times, and then I changed it for Perplexity.
However, even if Perplexity is good at giving answers to some specific questions and can understand Spanish (my native language), I still use my r1 when I have some questions or requests.
I have to ask in English and receive a reply in English. I have to carry it along with my phone. I know I can ask Perplexity directly in my phone. But having a dedicated device that looks so awesome and orange makes me want to use it more often than my phone. Also, the Magic Camera! That’s a killer feature.
This is important. ChatGPT is great at getting what I say even when I switch quickly between speaking Swedish or English. And it can help me with things like expressing myself in English.
A thing I would love to use the r1 for is to have it quickly translate Swedish to Norwegian for me and place the result on the clipboard of my computer or phone. With the right API:s to the device I could create an app that does this. Where the r1 would take text from the app, or (at my choice) start to listen for me speaking and then use that as input.