Today’s email attached. I get it - I can kind of imagine what the hard-ass working team is going through right now. I’ve been on those marches - they can crush a soul.
None of these are valuable updates. None of these address anything that has anything to do with a LAM. Rabbit Hole updates? You can now see your history. No additional integration points, not even a sniff of teach mode, not even fixing the existing number of integrations i can count on one hand.
Is nobody paying attention to this? I keep seeing people begging for their later batch releases, and I feel bad. First person to respond who can show me their order confirmation can have mine so you can cancel. Something tells me you can’t cancel…but i hope you can. You just have to pay shipping - and like $10 for a couple of beers.
Most exciting part of the update is the animated bunny jumping and eating.
I feel like by this point, if there was actually an answer to my questions about how this could ever possibly work, the founders are laughing because up their sleeves resides the most amazing tech (again, and scary), that I’ve ever seen. If they don’t have the answers, they are beating up their poor engineering staff who knows it can’t be done the way they’ve sold it. Would attach the email, but can’t do pdf’s here and I’ve already wasted enough time thinking about this and kinda don’t want to cut a screen video or a bunch of screenshots. Sorry. Trust me, it stinks.
I’ve seen people cancel. I am not sure what your issue and disappointment is.
You have to wait for LAM? As Jessie announced and said you can cancel at release?
I’ve likes the updates myself. I didn’t care for the rabbit animation, it got me nothing but I did really like that I got to see long form displays of my text and I could now direct it to use WoframAlpha or Perplexity if I want an information source.
Nothing you mention has anything to do with an action model. You are crediting the company that claims that can interact with any application, written on any platform, on any device, for allowing you to oustsource a prompt to another provider. You could do that on 3.5 with plugins.
Sounds like you don’t know what this device has actually been sold as.
I saw what was in the CES Video, thats to be coming… I didn’t set my expectations that high on a NEW kind of Device. I suspect hard work can make those possible.
If you saw the quarter release video you’d know where Rabbit was in at in this and not have these expectations either, you’d know what to expect and when to expect it. Though like others on discord and here just to complain. I don’t think you paid attention or could grasp it. They have been transparent, LAM is coming as stated, what else is expected. I don’t have any idea. The Memory coming today an quick AI access was at my low bar of expectations. I am in the show me state and want to be shown to believe it. Its cool though so far. I am excited to see more to come.
people can cancel and refund and all that. There have been a few cases of people having issues but that happens with every company. As for the features and fixes… I do believe they are bringing them as fast as possible for the team size they have while adding little features that are low hanging fruit to keep users engaged. Much of this is so much harder than people realize, LLMs are not easy to tame becuase they do weird unpredictable stuff at times. The community is also extending the device just look at LAMatHome which is a fantastic project. There are frustrating things going on but also some bright spots and a hopeful future.
I understand you are upset with the trajectory of the progress but you have more than enough options if you are unhappy including a 30 day return window. So I think if you don’t like the device you should just return it.
Can’t grasp it? Grasp what? I’m happy to put my nerd quals up against anyone in this space. You have still yet to answer my very simple question - how can a LAM integrate and work at scale with technologies that are purposely built to not be integrated with or scaled.
Sure - all of this can be seen as complaints - which is a valid point of view to have given the trajectory of this project. The “complaining” didn’t start until I started getting countless responses like yours. Which don’t address any of the concerns I have with the viability of this technology - because I know how these things work. I’ve built them. I’ve had a hand in building/architecting several of the applications/sites/services that they have used for examples in the past. Like built them. So I kind of do understand what is needed to be done to make this a reality.
I went into this eyes wide open - I knew it was a roll of the dice. I don’t want a refund. I want someone to explain to me how any of this is ever going to work in the foreseeable future - and even if it could, the price of the device would pale in comparison to the carrying cost of the cloud infrastructure that would have to exist to make even part of this a reality.
Again, offering my device to someone on the list so that they can have it, and hopefully get joy out of it without paying full-price. $10 for beer and pay shipping, that’s my only ask.
Yeah - let’s see that demo. I’m happy to share my GPT → Alexa via AI talking to itself demo. LAMatHome is fine - but there are a million GIT projects that do the same thing. Anyone with coding skills and free time can create their own LAM (I call mine the TAM (tiny action model).
My point is that you can do that with pretty much any GenAI tech. You just have to do a crazy amount of work (beyond the actual integrations) to configure it device by device, and it is limited in its scope of what you can hit directly through your own code (but a lot of this can be fixed with even more custom home automation projects).
The entirety of this is a workaround to do what the device is supposed to do itself. Use a poorly made LLM to do your own LAM. Might be fun, but kind of not what most people who bought are signing up for.
TBH, putting RAG on your conversation history is actually a fairly useful addition, especially if it can search transcripts of voice notes. I’ve also found that it’s slightly better at answering questions now than it was at launch, the article view that it gives for longer answers helps a lot here.
That isn’t to say that they don’t need to get their act together with the LAM functionality, what was released (even knowing that Teach mode wouldn’t be available at launch) is not what was promised. The entire benefit of having an AI control your apps is that it should be able to handle some flexibility, covering edge cases dynamically. For example, it should be able to do the majority of tasks available in Spotify, like adding a song to a playlist, playing a custom playlist, or other non-standard playback actions. The r1 simply cannot do that, it’s clear that the LLM is just calling prewritten UI interaction scripts for global search and playback on request, nothing more.
And while this week’s update is actually fairly useful, I’d like to know who greenlit Magic Camera this soon in the roadmap, it feels like it was the marketing department, especially with the hidden rabbits everywhere. It’s actually a fairly fun feature, but you release those kinds of features once you have the fundamentals handled, and preferably in a way that benefits the customer, rather than just trying to get social media engagement/brand awareness, which is what it feels like they were trying to do.
So that’s the thing. I actually don’t believe that teach mode is possible in most cases, realistic in all cases, and not even close to being scalable.
I think your analysis is fantastic - I do think people will find joy and use of this thing - it’s cute, it can be fun, it’s as annoying at times as every other LLM model. But once that wears off, it might as well be a cute orange brick. Without LAM - which I still will argue is not going to happen, what becomes the point of this thing beyond novelty, or certain people’s affinity for a device that works for them.
Just once - just one single time, would I like someone from Rabbit or from this group who knows their stuff, to describe how this could ever be a reality. With details. Not just parroting the normal lines - it will work in the next release or it is in beta or it’s hard but it will be solved.
Out of the infinite integrations it has claimed that it can do, the velocity of any LAM improvement, and the issues you rightly point out with the rudimentary actions it can do, should probably make most people pause and take a deeper think.
Even if you could get learning mode right, it wouldn’t work on native apps, and it would put the onus on the user themselves to update it every single time a site changes a webpage. Which happens about once every two weeks. Don’t get me started on two-factor authentication. Do you think highly trafficked sites and services want bots keeping open sessions with them?
I just want to learn what I’m missing. Someone, please take me to school so I can shift my thinking.
If LAM is actually a reality, it would be the most valuable technology ever created in my opinion. It makes AI into a QA Testing Automation machine that can infinitely scale. It is the coolest thing that could ever happen, and also the scariest. If the flash crash was bad - just wait for that level of crazy…
Me also very curious about LAM. one of the reasons I bought this thing. So the Rabbit could handle all my social media posts. Which would be such great thing!
“Hey Rabbit, please post my last picture on gallery to my instagram and tag Robbert Tagger and Philippe Master please. # this will work and add a nice prompt describing the picture”
Also would love to see my Rabbit take an incoming invoice, but it into my accounts flow and pay it, without me doing anything.
I was actually amazed that the Rabbit can’t detect my voice or face. If my girlfriend speaks to it it’s the same as I’m speaking to it. It still addresses to anybody as Vincent.
My hopes are very high and can’t wait for this LAM teaching mode.
So let’s unpack the that first use case in terms of what the Rabbit would have to do to make your vision (which is a great use case) a reality:
(1) It would have to translate what you (or your gf lol) says into a sequence of commands or actions - and then iterate through them to perform your task. That’s what a LAM is supposed to do right?
(1.5) It would have to login to Instagram from wherever it may be trying to do this. Even if it kept an open session to your instagram, those timeout. So let’s say it keeps your credentials and can actually do that for you (scary). Those sessions time out - which means it has to login again. Probably from another IP Instagram has never seen before, so Instagram will force two-factor authentication. Rabbit obviously can’t do that - you can barely enter txt. But even if you could, you’d have to grab your other device(s) to get the info you need to pass two factor auth, or approve it from another device.
(2) It has to find your latest picture from a gallery. Since the native iOS instagram app and the android app can’t live in a virtual version or emulation on a cloud server in the Rabbit Hole, it would have to use the insta website. The insta website, which would be running on a virtual server somewhere in the world, wouldn’t have access to your gallery - so it would have nothing to post. Even if you could emulate native apps in the cloud and teach them, same problem - they would be running on another device somewhere else that isn’t yours - hence no images to post.
(3) Then it would have to post the image and do the tagging. Same problems as above, and the API sets that could have allowed this (at a huge cost for integration to you or Rabbit) don’t exist any longer, and wouldn’t scale if they did.
This is the root problem of the LAM. It has to emulate human behavior across multiple devices that aren’t yours. Sites/services aren’t a big fan of headless requests coming at them from random IP’s, nor are the content servers that sit in between.
Even if all of this worked seamlessly - which I can’t see how that could happen - but let’s assume it could. If you teach an automation to do something, the second that pretty much anything on the website you are teaching changes, you have to re-teach it. Automated web interactions are absolutely possible - people have been doing them for years and years to test sites and automate QA and other things. But they are brittle, horribly difficult to setup, and even harder to maintain over time.
Same. Flushed a lot of money down the proverbial toilet, and an almost incomprehensible amount of my own time trying to solve a lot of this stuff myself. It’s maddening. TBH - I’m pretty sure within the next year Siri is going to solve my #1 pain in the ass issue - controlling the chaos that is my own “smart” home without having to scream at alexa all over my house lol
So if I read your posts correctly, you bought an R1, are unhappy and want to unload it now after a very short period of time using it, are contending that what they are saying they will be doing is “impossible”. As for what you’d like, you want them to tell you how they are going to do it, because you don’t see how it could be done (the LAM aspect of things). So to summarize that, you want them to share details of their proprietary information in public (thereby making said info public domain and valueless), you want the future features (yes, it was always advised to be a future feature) NOW, but you acknowledge that it will be a very high level programming issue and take time…but you want it now and are unprepared to wait for it. BUT you want to know exactly how they are going to do it, which basically blows their entire proprietary model. MMmmmmkay. You state you “can’t see how it could happen” but yet aren’t ok waiting to find out. Seems a bit of strange logic you are going through here. Your thinking of how the automation works is rather basic. The website changing has little to nothing to do with API permissions and login credentials, otherwise other things like Alexa, Google Home and Siri wouldn’t work for more than a day before having to be re-taught, which isn’t the case. There will be security issues initially, and there will be companies that will block attempts to automatically perform actions. But over time, this will change, as this is a direction that is new and requires time for everyone to settle into a new normal.
If you want to get after it, I will. I guarantee you are going to have to rely on superlatives and gerneralizations and try to confuse me with archane tech details. I’m in - you have absolytely missed the point here. And I’m happy to go through line by line everything you put together - and will continue to do so until one of us just gives up. You never know, maybe this is the path for me finally seeing something that I don’t beleive exists.
transparent, LAM is coming as stated, what else is expected. I don’t have any idea. The Memory coming today an quick AI access was at my low bar of expectations. I am in the show me state and want to be shown to believe it. Its cool though so far. I am excited to see more to come.
