That’s the same case for composing a mail with gmail. it couldn’t find the textbox for the body of the mail. so it ended up sending a mail with just a topic title.
It seems like LAM works with visual information… I hope it is a combination of both visual and webpage code but i’m not sure.
Hope it gets a fix soon. It keeps using the doodle page as a starting point and starts exploring it instead of visiting pages mentioned in the prompt.
I got the doodle page once now. More concerning: it had remembered my google credentials after I deleted the session data in the profile privacy settings.
I must confess that even in a classic web browser this button is a headache for my mouse (talking about mail.com) 
Hi again, I managed to place an UberEats order today using the LAM web interface!!. I was able to login by helping the rabbits to solve the puzzle. However, when I tried doing the same with the LAM R1 interface, the puzzle became unsolvable, showing dice addition problems where the images don’t match the expected results. I even used a calculator to speed things up, but there is nothing to do when the correct answers are not proposed in the captchas. On the LAM web interface, once I log in, UberEats stays connected, allowing me to continue interacting with the rabbits without needing to log in again. It would be great to have this same feature in the R1 interface as well so I can play with the voice command and maybe pass a new order again!
I’ve finally logged in using the LAM r1 interface, and it’s amazing what the rabbits can do. For example, they can search for the price of a Coca-Cola at the nearest subway station, retrieve details from my last order, including the date, cost, and contents, and even reload that order to create a new one. It also allows for adding a tip afterward and managing delivery priority. 
Hi guys,
Short question here, I am not sure to understand the final purpose of the LAM playground.
Is it a one shot purpose, meaning I show him how to perform something till the end, but then ? I just hit pause and quit the page?
How to recall what I have learned him to do from the R1 then?
Exemple with amazon.com.
First from the rabbit hole I will show hom how to connect to my account and then show him for exemple how to search and article an add it to my cart.
Then how to proceed on the R1 itself?
I just press to talk and have to say what? « Please go to amazon.com and search for this article and add to my cart the greatest reviewed one »
Is that correct?
If so, it could also have been nice during the playground part to have above the description field a kind of « command to trigger the action » no? Like when I am asking you « Find this article and add it to my Amazon cart » he knows what to do.
I am lost with the purpose, it is for one shot or to recall actions learnt? If so why not having a kind of trigger setting to recall the actions.
Thanks for the clarification guys 
That’s something for teachmode! I believe this is more of like a preview of what the LAM system is capable of at the moment.
Go to XXXXX website and claim all the rewards for today 
If we compare it to typical LLM behavior:
Teach Mode: One-shot prompting. You show it how to do a task once, then ask it to do the task again later. The only thing it “remembers” is what you showed it during the lesson.
LAM Playground: Zero-shot prompting. You ask it to do something random on the internet for you, and it does it’s best to comply. The only thing it “remembers” is what gets written in the rabbithole.
Both: Use cookies from cookie jar
I today had the bright idea of trying to use LAM Playground to access the Rabbit Hole to find information the R1 cant remember as it was not saved in a note but in conversation. However the Hole did not allow the LAM Playground to log in as it is not human.
Yes, Rabbit SSO would be useful
Hi @clairenicholson have you solved?
… if not my workaround to turning around the issue could be to save your favourites prompts in a drive like Google, OneDrive, notes, websites where you can paste something like pastebin or maybe also a github Repository (public and/or private depends on your needs), after paste your prompt try to use the LAM to visit this document and you will see the rabbit using this file/document as really a live prompt!
I am working to create a “multilanguage web directory” like this on github, that remain public for all the prompts, but simply forking the project and put it private, users can add some personal/private prompts
(if anyone wants to contribute, please contact me here via DM)