The adorable, whimsical and a very orange Rabbit R1 is here and it left us wondering if these AI-powered gadgets are even possible. The most intriguing tech in the R1 is what Rabbit calls the “Large Action Model”. Where a large language model, or LLM, is all about analyzing and creating text, the LAM is supposed to be about doing stuff. But in our week-long testing, the R1 has consistently proven to be underwhelming, underpowered and undercooked.
Category
🤖
TechTranscript
00:00What is this?
00:02You are holding a taco.
00:05I think AI gadgets might just be impossible.
00:08I've been testing this thing, the Rabbit R1, for about a week now,
00:12and I don't know how to draw any other conclusions about it than that.
00:15The R1 has the same big ideas as the Humane AI pin, the smart glasses from Meta and Ray-Ban, and frankly,
00:22countless other apps, platforms, and gadgets, to let you do stuff just by asking. In theory,
00:27you can just spend your day living your life, walking around, and chatting away with your gadget.
00:32What does this plaque say? Do you know what monument this is? What kind of animal is this?
00:37It all seems so easy. It seems simple. It seems kind of futuristic.
00:41It just absolutely, 100%, does not work well enough to be worth your time or your money.
00:47Were you recording on that?
00:49It was so wrong!
00:51The plaque in the image contains an inscription that reads,
00:55These in the presence of Washington at Valley Forge, whose faith in the continuing union inspired their hearts,
01:01and whose light in illuminating their minds was as the light which led,
01:05these made the arsenal of democracy and the arsenal of freedom.
01:10Again, it's like, the vibe is the same. All the words are wrong.
01:15Basically what I'm saying is, life with the R1 should be awesome, but it's kind of a mess.
01:21It's kind of a mess.
01:23I cannot, I cannot, but cannot, I cannot, I cannot, I cannot, I cannot.
01:28Can I assist with something else?
01:30With the R1 in particular, I think I can probably explain most of its issues just by talking about music.
01:36The R1 connects to Spotify, which I like. I use Spotify. This seems like a win.
01:41I love the idea of a dedicated music gadget that I can take on walks or runs, use for podcasts, all that good stuff.
01:48A Spotify iPod is actually a thing I've wanted for a long time.
01:52And it's, well, let's just try it.
01:54Play Fortnite by Taylor Swift.
02:07That is Fortnite by Taylor Swift. You'll love to see it.
02:10Play my Discover Weekly playlist.
02:19This one's my favorite. It's a song called Can You Discover by a band called Discovery.
02:26Which is dead wrong, but you can see how it got there, right?
02:32Anyway, this is not a Spotify iPod. It's just kind of a disaster.
02:37I can't even really figure out how to control the music either.
02:41Sometimes I press the button to pause and nothing happens.
02:44And if I scroll down, it goes to the next song.
02:47But if I scroll up first, it goes to the home screen.
02:50And then I have to click back in just to be able to click again to pause.
02:53It's all just confusing.
02:55No, don't play the... Now it won't even pause.
02:58This is the button you press to pause the music.
03:00Okay, we did it. We paused it, everybody. Great job.
03:05The R1 connects to Spotify through what's known as a large action model.
03:09Which Rabbit describes as basically a way for the device to use your apps.
03:13Actually navigating their interfaces for you.
03:16The large action model is really Rabbit's biggest idea.
03:20That instead of just connecting you to information, answering questions, all the chat GPT stuff,
03:25the action model can actually do things on your behalf.
03:29On the R1, it only connects to four. Just four.
03:33Uber for rides, DoorDash for food, Midjourney for generating images, and Spotify for music.
03:39But in theory, in theory, you'll someday be able to teach the R1 to use any app for you.
03:45But there's theory and there's reality.
03:48I have never once, for instance, successfully managed to order DoorDash.
03:52Order food.
03:53DoorDash may take a while to load on Rabbit OS and may not be available in all regions.
03:58There was an issue with the food ordering service.
04:01Please try again later. Thank you for your understanding.
04:04And then it fails.
04:05Same with Uber, though I at least successfully got a little further in the process a few times.
04:11Get me a ride. I'm going to Dulles Airport in Washington, D.C.
04:15Uber may take a while to load on Rabbit OS and may not be available in all regions.
04:20Apologies, the Uber ride request couldn't be processed.
04:24Please provide your drop-off location.
04:26I did that already. You get the idea.
04:29Also, I'm sorry, but if you have to tell me how slow an app is while it's loading,
04:34you maybe shouldn't have shipped that app.
04:36Anyway, so far the large action model is basically a bust.
04:40At best, it's super hacky, as you can see.
04:43And at worst, it's not even actually novel AI at all.
04:47There's plenty of software out there that uses computer vision to navigate apps,
04:51and it actually sounds like Rabbit is using one of them, called Playwright.
04:54Let's just ignore all of those features for now.
04:57Neat sounding idea, total mess in execution.
05:01And unfortunately, that's not where the mess ends.
05:04Like with the Humane AI pin, there is an incredibly long list of really basic things
05:09that this device just can't do.
05:11Remind me in three hours to take out the laundry.
05:15Laundry reminder saved to the rabbit hole.
05:18It's a note, but it won't remind you.
05:21Set a timer for 45 minutes.
05:23I cannot set timers. If you need a reminder.
05:27Send me an email with five recommendations for things to do in Washington, D.C.
05:33Visit the National Mall, a linear park filled with monuments,
05:36such as the Lincoln Memorial and the Washington...
05:39Can you send me all this in an email?
05:41I cannot send emails.
05:43You can save this information by taking a note in the rabbit hole.
05:47And of course, since this is an all AI gadget, you can't do anything at all offline.
05:53You can add a SIM card in here, like I did, so you can get data everywhere,
05:57but it doesn't have GPS or navigation, and it can't make calls or send messages,
06:02so it's really not more useful than whatever device you're already inevitably bringing with you.
06:07I do have some slight good news here.
06:10A couple of days ago, actually, while we were in the middle of shooting this video,
06:13Rabbit updated the R1 and fixed a bunch of what I would call show-stopping bugs.
06:18The battery no longer drains in a few hours, even if you're not using it,
06:22though it's still not gonna last all day.
06:24The thing actually shows the correct time, which it didn't before.
06:27What a concept.
06:28Rabbit has said all along that it knows it's early,
06:31and will update this device frequently and quickly,
06:34and it's definitely true that a lot of what's wrong here is software.
06:37But even after this first update, we are nowhere near this feeling like a functional device.
06:42The main two things people will do with the R1,
06:45other than getting really mad at the broken app control stuff, anyway,
06:48is ask it questions and have it identify things in the world around you.
06:52The R1 does both of these reasonably well.
06:55I like pointing the thing at a sign or a plaque or a document and having it summarize things for me.
07:00Actually, we're in Washington, D.C., home of a million landmarks and plaques and touristy things.
07:05Let's go try a few, see how it works.
07:07The image appears to depict a large statue of Abraham Lincoln,
07:10seated in the Lincoln Memorial in Washington, D.C.
07:13There we go.
07:14This appears to be the Gettysburg Address.
07:19It handles all the standard chat GPT style stuff well enough,
07:23which isn't surprising, it's connecting to all that stuff.
07:25But it's also not that hard to get that stuff right.
07:28And it does tend to be really wordy when it does so.
07:31Plus, sometimes it works.
07:34What's the score of the series now?
07:36The Denver Nuggets won the series against the Los Angeles Lakers
07:39with a final score of 108-106 in Game 5,
07:44which took place on April 29, 2024.
07:47Jamal Murray was the standout player for the Nuggets,
07:50scoring 32 points and hitting the game-winning shot with 3.6 seconds left on the clock.
07:56LeBron James had a strong performance for the Lakers with 30 points and 11 assists,
08:01but it wasn't enough to secure the win.
08:04The Nuggets will now face the Minnesota Timberwolves in the second round of the playoffs,
08:08starting next weekend in Denver.
08:10And sometimes it doesn't.
08:12How do I get to the Lincoln Memorial from here?
08:15The memorial will remain open to the public during the renovation,
08:19with only a small visitor space located below ground being closed.
08:23Give me a moment to look that up for you.
08:26To get to the Lincoln Memorial from Homefront Dr., Washington, D.C., 20037, USA,
08:32you can take the blue-orange-silver line to the Foggy Bottom Metro Station,
08:36which is about a 15-20 minute walk to the memorial.
08:40The correct answer was go up those stairs right there in front of you.
08:43Alternatively, you can take the blue-orange-silver line to the Smithsonian Metro Station.
08:47When it works best is when you have really identifiable things.
08:50But at that point, do you really need the help?
08:52This thing can tell me what's on a can or what the label says on a bag of chips,
08:57but it says it right there.
08:59The Doritos Cool Ranch chips in the image show that a serving size contains 150 calories.
09:05So the number of calories in a serving of these Doritos chips is 150 calories.
09:10What are we accomplishing here?
09:12And honestly, what are we doing that I couldn't do with Google Lens on my phone?
09:16I think the answer is not that much.
09:18The Humane AI Pin, I actually understand in theory.
09:21It's a wearable thing that sits on your body, making some basic functionality more accessible.
09:27This is just another thing to carry in your pocket or purse
09:30that can't do anything your phone can't do better.
09:33I cannot for the life of me come up with one thing that this does better than your phone.
09:37And that feels like a problem.
09:39But let's just for one second leave all of that aside.
09:42I do believe the AI stuff will get better eventually.
09:45So let's just say that the large action model works.
09:48Again, it doesn't. I cannot stress that enough.
09:51But let's just say that someday it does.
09:53The question left then is whether the R1 is a good device for that future.
09:58And the answer is kind of.
10:00There is something about the R1 I find very charming.
10:03It's not fancy, as you can obviously see, but it's also $200 and doesn't have a subscription.
10:08So I don't expect it to be as nicely made as something like the AI Pin.
10:12And it's not. But I really like this orange color.
10:15I think it's a nice size.
10:17Purely as a little design-y gizmo, I actually dig it.
10:20And honestly, that does count for something.
10:22But nothing about using this device feels intuitive to me.
10:26You interact with the R1 in a couple of ways.
10:29There's this button on the side for activating it so you can talk to it.
10:32Look, the ears perk up when you do it.
10:34It's so charming.
10:35There's also this scroll wheel, which lets you...
10:37Thank you for the compliment.
10:39You're welcome.
10:40There's also this scroll wheel, which lets you page through menus and information.
10:44And there's the 2.88-inch screen, which is not a good screen, but it's fine.
10:50The combo of these things feels right.
10:52I really wish the Humane Pin had a screen for when you want to see the weather rather than having it read to you.
10:58Or you want to review something you've said before you send it off somewhere else.
11:02Screen plus microphone plus speaker plus camera feels right.
11:07And yes, I'm very aware that I just described a smartphone.
11:11And I think this thing is probably too close to a smartphone to have much of a chance to be different.
11:17But let's just keep going with this design just for one more second.
11:20Here's where it gets weird.
11:21This screen is a touchscreen, but only sometimes.
11:25When the keyboard shows up, you can type on it, which is good.
11:29But when you're in settings or looking at the weather or whatever, all you can do is scroll the wheel.
11:36This doesn't make any sense to me at all.
11:39And it makes the R1 harder to use because you have to hit the button with one hand and then sort of shift around and scroll with the other.
11:47There are lots of little weird things like this on the R1.
11:51You shake the device twice to get to the settings menu, but then there's no obvious way to get back.
11:56You just have to scroll around until you stumble on the back button at the top and then click the side button to go home.
12:03There's nothing indicating how that works and none of it feels at all intuitive.
12:08Honestly, the best single feature I found on this device by a mile is the voice recorder.
12:13As a reporter, I probably use voice recorders more than most.
12:17But this one is pretty easy to use.
12:19It does some AI summarization and actually has a pretty high quality microphone.
12:23You just say, start the voice recorder.
12:27Start the voice recorder.
12:31And there it goes.
12:34Voice recorder.
12:37Then you scroll down and hit stop because the touch screen doesn't work.
12:41Starting the recording now.
12:44Which it did after it was done.
12:47God, you suck.
12:49Oh boy.
12:50Okay.
12:51Anyway, the more I use this thing, the more I feel like Rabbit spent all of its time making the R1 fun and silly and whimsical.
12:58And none of its time making it work.
13:01I love this little bouncing logo.
13:03I like that it gets headphones when you're listening to music.
13:05But I don't care about any of that because it doesn't play the songs it's supposed to.
13:09Humane, it felt like, tried to mask its device's flaws under this really shiny, luxurious, thoughtful piece of hardware.
13:16And Rabbit went all the way the other way and seems to just hope you'll think this thing is a toy and not mind that it's broken.
13:24Right now, if you buy one of these, you're buying a fun-looking $200 desk toy and not a useful gadget.
13:30Because this isn't a useful gadget.
13:32I don't think it'll be one anytime soon, if ever.
13:35I do think the large action model is a cool idea.
13:38And it might be the future of AI in some meaningful ways.
13:42But there's not one shred of evidence in this device that Rabbit is going to be the company to figure it out.
13:48I continue to think there will be lots of interesting AI gadgets in the future.
13:52But for now, and probably for a while, your phone is just going to be better at everything.
13:58You don't need the Rabbit R1. Nobody does.
14:00Just keep your phone charged. You'll be fine.
14:03Are you saying I can't eat tomatoes?
14:06No, I'm not saying you can't eat tomatoes.
14:09This appears to be a red bell pepper, not a tomato.
14:13You are holding a red bell-shaped vegetable with the word cone printed on it.