In fact that is what it does the most of the times, at least in my unit.
1 Like
This is actually already in place, the prompt âbrowse [website] and do xyzâ should always start a playground session from r1.
3 Likes
The fact is sometimes with that sequence it tries to rely on Rabbit Hole or on the multistep reasoning AI search and I had to reset the memory, or at worse restart the device, because it repeatedly told me itâs not able to directly control websites.
I think some bug fixes should be taken.
An other thing Iâd like to see, @Simon, is the ability to create routines. For example: at a command like âCatch me upâ it goes to the gmail inbox and summarize e-mails, then goes to read the news, gives me an update of whatâs going on my Facebook and so on.
Or as it was showed in the demo before the launch, at the youtube advertisement, the command âStart my morning routineâ should trigger breakfast ordering, turn by turn directions to the nearest coffeeshop etc. Each command should have custom actions programmable by the user, so I donât need to instruct LAM to do a thing I do periodically.
2 Likes
I understand, but still come across situations like this:
Me: 'browse gmail.com and read me my latest email"
R1: I am unable to access or read your emails directly from gmail.
This happens with different requests too and it replies things like: Iâm unable to browse websites or interact with content on platforms directly.
I use the keyword âbrowseâ and my requests are pretty clear.
1 Like
Would it be possible to get the âr1â mode as the default instead of the âwebâ mode when calling LAM?
I think that would be much better, especially since the connection to r1 would then be stronger and it would also be much more convenient to just talk to r1 than to type something into some field.
The prerequisite is, of course, that it starts correctly; unfortunately, at the moment it doesnât seem to be running very stable.
1 Like
As I was writing before, when it does it I try to repeat the command, to refresh the memory by pressing 5 times the Push to talk button, or at worse, to restart the device.
1 Like
Yes⌠but⌠thatâs a problem right? It is unclear how to 100% start a playground session and it is really bad UX if the rabbit tells you it can not do something it has done before.
Rabbit still uses some of its tools wrong. Saving notes randomly or âlooking for placesâ⌠Keywords are at least clear and provide the user with some control to use the right tools. Like with the âbeta rabbitâ
2 Likes
Odd. I have never had to use the 5xppt yet. I do quite frequently delete the playground session data in the profile setttings if it stops working / gets lost.
1 Like
I have 2 priority wishes for the playground:
- recognise when a new tab gets opened and move focus to the new tab when it does. There are currently a number of sites where it gets stuck repeatingly opening new browser tabs.
- Improve recognition / interaction with frames on sites. Most notable example is gmail new email dialog, but also buttons where it says it is âclickingâ but the browser does not respond.
1 Like
Relevant to this, in my opinion, I think it would be nice if it is stuck (for my case it keeps on opening a new tab or is just stopping at the page while struggling to find the next best way to move forward), it could pause for a moment and ask us what to do next (such as a subtle vibrate or a sound through R1). This in turn could make it learn from us on what to do next time whenever it is in such a situation again.
1 Like
This is great! loving the R1 vibess 
2 Likes
I really like the way this system tells about its actions in a more conversational way.
https://www.reddit.com/r/singularity/s/Dd13XuBdCw
1 Like
Hallo to everyone!
I admit I didnât read all the comments below so maybe what Iâm going to ask is redundant.
I m trying to use LAM but it have some problems :I mean the Rabbit take 4ever b4 reach the site
1 Like
Sorry,anybody could explain me how 2 use the LAM playground?
Do I need a laptop?
I tried to ask my R1 to search for cheaper flight to Martinica using Skyscanner but it get stuck with the privacy options.
I tried something simpler as to know a link to read a certain manga and,again,it stuck.
I ask to R1 what I should do to make it work but it gave me just stupid answersâŚ
If anyone is willing to explain me how to use it in a dummy proof way I would be really grateful
1 Like
I really love the new LAM playground and it works surprisingly smooth in some of my test cases. Whatâs really annoying, is the trouble LAM has with cookie banners - but thatâs not LAMâs faultâŚ
1 Like
Has anyone solved the Amazonâs password requirement each time I try to purchase something? LAM gets confused on checkout, because Amazon is prompting it to enter the password to login.
1 Like
Hi @Meth ,
the LAM playground is on his first stage.
I love the LAM playground.
The performance will be improved in the next updates. It is in the focus.
In the following thread @jesse mentioned this:
1 Like
It doesnât work reliably for example with youtube and gmail. It hallucinates that it is unable to do certain things.
1 Like
It started behaving like that this morning for me.
1 Like
Alright, then you understand what I mean now. I think a keyword might work⌠âBetarabbitâ was at least clear.
I understand we want the Rabbit to be intelligent enough to pick appropriate tools by itself, but that is not always the case right now.
2 Likes