(WIKI) General Uses for the LAM Playground

LAM Playground: Unlock the Power of AI Automation

The LAM Playground gives users the ability to automate tasks across websites, applications, and more, using the power of Large Action Models (LAMs). Whether you’re filling out forms, navigating complex web processes, or managing social media, LAM Playground allows you to perform these tasks simply by giving natural language instructions. Below are practical use cases and a community table for sharing more experiments and tips.

How It Works:

  1. Voice or Text Prompts: Command the AI with instructions like “Add a movie to my Netflix watchlist” or “Find me the latest gaming deals.”
  2. Real-Time Execution: Watch the AI agent as it hops between websites to gather information or complete tasks.
  3. Cross-Platform Potential: Currently optimized for websites, but capable of expanding into apps, desktop software, and more.

Community Use Cases & Experiments

We encourage users to explore the capabilities of LAM Playground by experimenting with tasks and sharing their experiences. Below is a table where you can add your own use cases, complete with detailed steps and outcomes. This will help build a living library of ideas for all users to reference.

~Thanks @C28 for the idea for this wiki post creation.

Example Use Cases:

Verified Use Case Description Prompt Example Expected Outcome Source Notes
Add Movie to Netflix Watchlist “Add ‘The Matrix’ to my Netflix watchlist.” AI logs into Netflix, searches for the movie, and adds it to your watchlist. Check login credentials.
Automated Grocery List Creation “Search for ingredients for ‘spaghetti carbonara’ and make a list.” AI searches various cooking sites, compiles a list, and stores it. Make sure to specify serving size.
Book a Dinner Reservation on OpenTable “Book a table for two at The French Laundry for this Saturday at 7pm.” AI navigates to OpenTable, finds availability, and makes a reservation. Include dietary preferences.
Add an Event to Google Calendar “Schedule a meeting with my team on Monday at 10 AM.” AI creates a Google Calendar event and sends invites. Check time zone settings.
Set a Reminder in Todoist “Add a reminder to buy groceries tomorrow at 6 PM in Todoist.” AI logs into Todoist and sets up the reminder with the given time. Confirm reminder notifications.
Automated Flight Check-in “Check me in for my flight on American Airlines for tomorrow morning.” AI logs into the airline site, checks you in, and sends a confirmation email. Double-check flight details.
Monitor Price Drops on Amazon “Set a price alert for the ‘Apple AirPods’ and notify me when it drops under $150.” AI monitors the product and notifies when the price drops. Use accurate product listing.
Organize Spotify Playlists “Move all my ‘Liked’ songs from 2023 to a new playlist called ‘2023 Favorites’.” AI creates the playlist and organizes the songs accordingly. Check duplicate song entries.
Check top twitch streams online at the moment “Browse twitch.tv and tell me the top 10 streams online at the moment and what game they are playing” AI provides a list of the most popular streams online at the moment, what game they are playing and how many viewers they have It seems like it is not an accurate top 10
Make a CV/Resume “Browse kickresume website and make a resume for [name] whose main skills including x, y and z” AI creates a polished CV based on the skills identified
Verify If Currently Logged In “Go to <site> and check if I am currently logged in.” AI goes to site and looks for clues that you are indeed logged in, like personalized welcome, profile icons, etc. (or not logged in: a big “Login” button). I tend to do this now because I don’t want LAM to attempt 10 logins with [your_username_here] and [your_password] and make sysadmins mad.
Add Topic to Rabbit Community Forum “Go to community.rabbit.tech, create a new topic in the <category> category that says, “…”. Note at the end of the topic that this topic was written by LAM (Playground)” AI creates a new topic on community.rabbit.tech in the correct category, conveying the correct information and intention, and signing it by LAM Playground It seems tricky to get it to stop for review before submitting the topic.
Communicate with Healthcare Providers "Go to <healthcare site>, send an extremely polite message to my doctor about “…”. Or “go to <medical site> and check if my doctor has written me a recent message, if so, summarize that message for me (be discreet).” AI either sends a message to healthcare provider or summarizes and reads out a message from the provider. Use “extremely polite” or something similar when you want to sound nice and polite rather than terse and informal.
Creating a new event Add a new event titled ‘Team Meeting’ on <date> at <time>. Google Calendar
:x: Summarize unread Discord messages On discord.com, review all unread messages in all servers and/or channels (e.g., LAMatHome, omi, and rabbit), concisely summarize the most interesting and most official messages, then mark all messages as read. I want a summary of less than 200 words for all servers/channels put together. If all the most interesting messages mean you don’t mention some servers/channels, that is OK It methodically goes through each server and each channel and reads unread messages, scores them on interesting or officialness, then summarizes the top few messages to me in less than ~200 words. Discord.com I was not able to get this to work. It would look at one or two servers and one or two channels and then summarize for me. As far as I can tell, all servers still had unread messages.
Place Hold on Library Book “Go to <library site>, find the first book from the Dark Materials series by Phillip Pullman, find the most recent version of it in book format, and place a hold on it.” It looks up book 1 of the series, checks for formats, if multiple years of book format chooses most recent, places hold.
[Vision mode] Pay this Bill “Go to the website on this bill and find a way for me to pay as a guest.” It actually went to the correct website and told me the information I needed to enter in order to look up my bill. At a later stage, I would really love to say, “Pay this bill”, it takes a picture, goes to the website, enters the account number and last name from the bill, solves the CAPTCHA, sees the bill amount, and pays using my pre-populated method for that site (i.e., Credit card info).

Keep adding your use case here (Or adjust previous ones)! The more examples we gather, the better we can explore the full potential of LAM Playground and uncover exciting new possibilities. Let’s push the boundaries together!

6 Likes

None of them work as intended for me.

“Schedule a meeting with my team on Monday at 10 AM with Google calendar.” Starts an audio recording. :flushed:

I get that it is interesting to get a list of LAM usecases but it would be more interesting to collect actual prompts that work instead of ‘prompt examples’

2 Likes

Hey, thanks for your feedback! This post is actually a wiki, which means it’s meant for the community to collaboratively edit and add any value they find during their experiments. So if you find a working prompt or want to adjust any of the examples to reflect something that actually works, feel free to jump in and make those changes!

The goal is to build a living resource that evolves as people test out different use cases, tweak prompts, and share what they find successful. It’s definitely a work in progress, so if you’ve got working prompts or other ideas, please add them! Your input would be super helpful for everyone!



1 Like

Same here. This list suggests that these are working prompts. I actually became a bit frustrated that I couldn’t reproduce these outcomes. It literally says: “Below is a table where you can add your own use cases, complete with detailed steps and outcomes.” That suggests working prompts to me.

I fully agree with @Designer here. Examples of prompts should be fully working prompts, unless clearly mentioned otherwise. I like the idea of potential uses cases, but please clearly mark them as a potential use case.

2 Likes

I just wanted to point out that there’s actually a checkbox field called “Verified” in the table. This is where we can mark whether a prompt has been verified or not. The idea is for the community to try out different use cases and mark them as “checked” once they’ve been confirmed to work. So right now, a lot of these are potential use cases, and we’re relying on community members like you to help verify and refine them.

If you do find something that works, feel free to check the box or adjust the prompt—it’s all about collaboration here! I’ll make sure to clarify that these are examples and not all guaranteed to work yet.
Thanks for the input, your patience is appreciated as this list evolves!

3 Likes

Great Wiki, mate! Fantastic seeing this here

1 Like

Did tou write many of them?

1 Like

This would be super cool although I imagine the LAM would have to remain continuously open and refreshing the page to watch for the price change? There is probably a better more technical way though.

1 Like

This does already exists. It’s an extension for your browser and or app on your phone, called keepa. It also shows the history of the price.

https://keepa.com/#!

1 Like

Teach Mode should resolve this issue. R1 partially seems to work as a key phrase assistant, searching for the right action based on a word you use. Phrases are the key phrase in teach mode.

2 Likes

So you suggest we collect cool / ideal usecases?

2 Likes

Act as a LAM AI Agent expert in recognizing human usage of UI in webpage services, define exactly how verbally you command how to create a ‘insert your whatnots here’.

Hey Rabbit R1 Community!

Are you ready to unlock the true power of your Rabbit R1? It’s more than just a device—it’s your gateway to effortless, personalized web interactions. Today, we’re excited to launch an exciting and interactive challenge that will let you customize how you interact with your favorite web services. Let’s dive into the “Insert Your Whatnots Here” Challenge!

:star2: What’s the Challenge About? :star2:

This challenge is all about pushing the boundaries of how you use your Rabbit R1 to control web services via custom verbal commands. Whether you’re setting up an account, sending emails, or doing some online shopping, you can create your own unique, intuitive commands to streamline these tasks.

How to Participate:

  1. Pick a Web Service:

    • Choose a web service you frequently use (e.g., Gmail, Amazon, Netflix).
  2. Define Your Task:

    • Select a specific task (e.g., sending an email, adding items to your cart, searching for a movie).
  3. Create Your Custom Command:

    • Think through the necessary steps for that task and translate them into a simple verbal command. For example:
      • “Compose an email to [Contact Name] with the subject [Subject Line].”
      • “Search for [Product Name] on Amazon and add it to my cart.”
      • “Find and play [Movie Title] on Netflix.”
  4. Test and Tweak:

    • Try out your custom command with your Rabbit R1. If it doesn’t work perfectly, refine your wording and try again.
  5. Share Your Masterpiece:

    • Post your custom command in the comments, along with the steps you used. Let’s inspire each other with clever and creative ways to enhance our Rabbit R1 experience!

Example Scenario:

Web Service: Google Calendar
Task: Creating a new event
Custom Command: “Add a new event titled ‘Team Meeting’ on [Date] at [Time].”

Steps to Create:

  1. Open Google Calendar.
  2. Click ‘Create’.
  3. Type the event title “Team Meeting”.
  4. Set the date and time.
  5. Click ‘Save’.

Now it’s your turn! :star_struck:

Why Participate?

Not only will you get to explore the full potential of your Rabbit R1, but you’ll also be part of a vibrant community that’s constantly innovating. Whether you’re a tech whiz or just getting started, this challenge is a perfect way to experiment, have fun, and improve your Rabbit R1 experience.

:rabbit: Share Your Genius Commands:
Post your custom commands, ask questions, or brainstorm ideas below. Together, we can curate a community-driven guide to transform the way we interact with web services.

:rocket: Push the Limits!

We encourage you to get creative and think outside the box. Your commands should aim for efficiency, originality, and just a hint of genius. The most innovative ones will make it to our Weekly Showcase, where we celebrate the best Rabbit R1 use cases in our Community Hall of Fame. Will yours be featured?

Note for the Polyglots:

We’re constantly striving to make Rabbit R1 more inclusive, so if your preferred language is still not supported, don’t worry—it’s coming (slowly but surely!). For now, let’s focus on what’s possible, while continuing to remind the powers that be that a true multilingual Rabbit R1 would make all the difference. :earth_africa::wink:

Weekly Highlights & Future Archive:

Each week, we’ll feature the most inventive commands in our showcase. In time, we aim to build a legendary archive of user-generated commands that will inspire Rabbit R1 users for years to come.


#RabbitR1Community #CustomCommands #WebExperiences #TechChallenge #MoreLanguagesPlease

Can’t wait to see what you create! :globe_with_meridians::tada:

3 Likes

So for ages, I’ve been trying to work out how I can be on just my R1 - get it to do something - and get the information into somewhere useful, reliably - and I think I’ve got somewhere!

Idea 1:
Use a link shortner to put a longer link on there - and therefore make it much easier to say - e.g tinyurl.com/SOMETHING - this means the LAM actually seems to get there 100% of the time which is great!

Idea 2:
I do a lot of my client work on notion - so I can say something like: (after I logged into notion on the web LAM playground)
“Navigate to Notion AI, draft [thing], save to private pages”

Or…“have a conversation on…”

I have also had a bit more success by letting the LAM playground do a bit more of the guessing - so instead of “click the purple button that says XXX to do XXX” - but actually if you take a step back… it seems to do better… I have had some pretty consistent success asking it to review my calendar for x date - but shocking when i tried to be super specific with the buttons!

3 Likes

NOS Latest News (= news in the Dutch language) – Create a news summary in the Dutch language of the latest headlines
“Go to nos.nl/nieuws/laatste and write a 50-word summary for each of the first 10 stories in the Dutch language”

When you run this in r1 mode, the AI speaking actually Dutch sounds pretty good!

It actually gathers data from each of the underpinning news stories. It doesn’t stick to the word count, but it is still nice.