Creating a Custom Rabbit (Skills)

[Designer]

I was wondering what people think about the idea of using other modes. (We had ‘betarabbit’ ofcourse and I hope there will be some other official modes)

I believe it would be great to be able to create our own modes in the rabbithole.
I think it could be as simple as a custom GPT (and injecting your own instruction).

A example of a mode I would like:

Using the keyword “conversation rabbit” would trigger a mode for when you don’t want R1 to search the web but instead just respond in a more conversational, playful manner—like throwing in some creative flavour that is less focused on providing factual information.

I’ve seen people talk about wanting to play (roleplaying) games with their R1 but how cool would it be if you could save your idea in a mode available to trigger with a keyword.

Any ideas on modes you would create?

[C28]

This could be triggered with the name of the rabbit too!

“Convo Rabbit, let’s chat”
“Research Rabbit, let’s chat”
“Verbal Abuse Rabbit, let’s chat”
“Early Modern English Rabbit, let’s chat”

Then, the requested rabbit would speak with you until you request another rabbit model.

[PaulBacon]

I love this idea! And if you could switch between modes like @C28 suggests then even a simple “hi Miffy” could set it in the kids mode that was suggested here.

It also aligns with the idea of having multiple profiles to work with that was suggested here.

You could have a “George” Rabbit that always likes to work with imperial metrics, “Werner” who speaks German and uses a European localization (answering to the request for local metric systems), or “Ruth” who is set up to have deep and meaningful conversations. If we could get multiple voices a rabbit could have it’s own voice and if they would have their own memory section it would be like having a whole family of individual rabbits in your pocket to interact with.

These could be called “Rabbits”, the Rabbit Hole could be renamed “The Warren” and the integrations (or teach mode rabbits) could be renamed “Rabbit Holes”.

Cool idea!

[Designer]

Definitely great to be able to keep it in a mode. Giving it it’s own name on screen. And maybe even it’s own face/icon.

[Designer]

Thanks for bringing these ideas together! Yeah that was what I thinking. Sharing rabbits (or concepts for rabbits) would be something very interesting aswell.

[Designer]

Automatically detecting what mode should be used would be great too ofcourse.

[Designer]

I don’t think we should clutter the simple experience with all kinds of complicated things but I still like to drop this idea. (It might solve some challenges teachmode faces aswell)

I like the idea of different kinds of assistants with their own skills,
switching between these should be intuitive. You could look at these ‘Rabbits’ as forms of apps. (I know Rabbit doesn’t like apps but an assistant skills are also kind of like apps)

It would be nice to make them yourself and be able to see the ones you made, or make the rabbit recognize what kind of Rabbit/Mode is appropriate. Just like how the rabbit uses different tools at the moment.

[Designer]

@simon are there any plans for personalizing our assistant behaviour? Or a form of custom instruction making since it would add a lot of creative freedom to using the R1.

[Argo]

This falls in line with some of the design ideas I’ve written about on game design for the r1. Specifically…

@Argo

In a perfect world, we can create “modes” for the r1. When the r1 enters a “mode”, it restricts what types of actions and interactions you can ask of it. For example, I would love to create a log mode, where vision entries get saved under a flora, fauna, or mineral category. In addition landscape shots save under a “cartography/location” category. This mode essentially applies a fixed filter prompt that encourages the summarization of vision entries to fit these categories. This might even be a function of a greater r1 ARG companion mode (I haven’t really decided on a name…). This mode based design is primarily to limit false queries. With a simple “shake to exit”, you return to the base r1 functionality.

---

I never considered using it to alter the “personality” of the rabbit, but rather to help create fixed experiences. Especially if the full vision of the r1 is reached, having a lot of custom functions that the community makes might result unreliable prompts and false queries.

[Argo]

With these types of menus, I also proposed a design you might enjoy in my virtual pet games write up. The intention was to keep a simple menu interface, while still allowing for some nested menus and providing some depth to work with in UI. Essentially once a user scrolls onto the menu they would like to interact with, using the PTT could select that option, and rotating the device could act as a secondary input. Take a basic inventory menu for example. PTT might give a “simple” view of the item highlighted. It could expand the selected item to show a description and stats. Rotating the screen instead opens item options (equip, delete, craft, trade, sell, etc.)

@Argo

A big thing I wanted to add was some UI ideas I have. The r1 isn’t really intended for gaming, especially because we really only have a single button to use. This is one of the big restrictions we have to work around with the designing process. If we continue to avoid using the touch screen, we will have to find an analogue solution to this issue. With some creative menu choices, we can start to open how many options a user has to interact with the gameplay.

Once the user “opens” the game, I would expect the device to stay in this game mode until either: the user chooses an exit game button or the rabbit is reset (5 PTT presses, in the case of a failure in the program this avoids soft-locking users in “game mode”). Starting out, the users device is upright. I expect this to be some sort of pet selection menu, with a list of your owned pets to scroll through. You scroll with the scroll wheel, and select with the PTT button. We could start building multiple menu layers, but I would rather avoid layering too much and instead use one of the unique aspects of the device.

The r1 is square, no matter what way you hold it the overall form factor will remain the same. We can use the rotation of the device as a secondary menu select option. The PTT button could open a “stats” breakdown, and the rotation could transfer the player to a new screen. We can have 4 screens in total, each with their own menus or buttons to cycle through.

Screens 1,2, and 4 can be considered the standard “handheld” modes, while screen 3 can be reserved for more menu/setting intensive options. I say this mostly because screen 3 requires two rotations from the default position, which is quite awkward in standard use.

[Designer]

Yes, I thought about how flexible making and using a custom GPT is. It can be and do anything…

[simon]

@Designer

@simon are there any plans for personalizing our assistant behaviour? Or a form of custom instruction making since it would add a lot of creative freedom to using the R1.

It’s definitely something we’d like to look into, but it’s not (as far as I’m aware) something we have been able to spend any time scoping just yet.

[Designer]

Thanks for your reply. Good to know, it’s something you would like to look into. I think it could be really powerful to get the conmunity to create with the R1.

[Designer]

I have an watch app on my Galaxy watch called ‘power chat’ AI that provides the most simple way to add changes to the system prompt.

Why wouldn’t we do something like this for the R1? Everyone could experiment with their own preferences more.

ChatGPT also has this feature to add instructions and change assistant behavior.