Rabbit R1: Barely Reviewable

Marques Brownlee
29 Apr 202419:53

Summary

TLDRThe Rabbit R1 is a portable AI assistant device designed by Teenage Engineering, featuring a recognizable lightweight plastic cube with a built-in screen and a single button for interaction. It connects to the internet via a SIM card and offers quicker responses than the Humane Ai Pin. The device includes a scroll wheel for UI navigation and a swiveling camera for multimodal interactions, including identifying objects and summarizing text. Despite its innovative features, the Rabbit R1 has drawbacks such as poor battery life, lack of basic functionalities, and the absence of a touch screen for navigation. It is priced at $200 with no subscription fees, but requires a separate SIM card for cellular access. The device's 'large action model' concept aims to perform tasks within apps as a human would, but currently only supports four apps with limited success. The potential for future improvements with more training data and additional features like 'Teach Mode' is acknowledged, but not yet available. The review criticizes the trend of releasing incomplete tech products with the promise of future enhancements and advises consumers to consider the product's current capabilities rather than its potential.

Takeaways

  • 📦 The Rabbit R1 is a portable AI assistant device, designed by Teenage Engineering, featuring a recognizable, lightweight plastic cube form factor.
  • 🔴 It has a single button for interaction, a built-in screen, a SIM card tray, and a USB Type-C port for charging, with a notable absence of cellular built-in.
  • ⏱️ The device is quicker at answering questions compared to the Humane Ai Pin, although starting from a low benchmark.
  • 🌐 It relies heavily on cloud processing for its functionalities and can connect to headphones via Bluetooth or use its built-in speaker.
  • 🔄 The unique user interface includes a scroll wheel for navigation and a swiveling camera for multimodal interactions, including visual queries.
  • 📸 The camera can identify objects or read text from a screen, providing summaries of long emails or articles.
  • 🔋 Despite its many features, the Rabbit R1 has poor battery life, lasting only about four hours and taking 45 minutes to charge fully.
  • 🚫 The device lacks basic features expected from a modern assistant, such as setting alarms, timers, recording videos or photos, and email functionalities.
  • 🛍️ It is priced at $200 with no subscription fees, which is a significant drop from the cost of the Humane Pin, but still requires a separate SIM card for cellular use.
  • 📱 The Rabbit R1 comes in a minimalist package with no additional accessories, and its construction, while sturdy, is entirely plastic.
  • 📈 The device introduces the concept of a 'large action model,' which is intended to perform tasks within apps as a human would, but currently only supports a limited number of apps.

Q & A

  • What is the Rabbit R1?

    -The Rabbit R1 is a virtual assistant in a box, designed by Teenage Engineering, which is a lightweight plastic cube that can be carried around in a pocket like a smartphone.

  • How is the Rabbit R1 different from the Humane Ai Pin?

    -The Rabbit R1 is not wearable and has a built-in screen instead of a projector. It also has a SIM card tray and a USB Type-C port for charging, and it is reported to answer questions more quickly than the Humane Ai Pin.

  • What are the unique features of the Rabbit R1?

    -The Rabbit R1 has a scroll wheel for navigating the UI, a swiveling camera for multimodal interaction, and a large action model for performing tasks within apps.

  • What are some of the limitations of the Rabbit R1?

    -The Rabbit R1 has poor battery life, lacks basic features such as alarm and timer settings, video/photo recording, email sending, and calendar functionality. It also can provide incorrect answers due to AI hallucination.

  • How does the Rabbit R1 interact with other apps?

    -The Rabbit R1 has a 'large action model' that theoretically allows it to interact with apps in a way a human would, using keyboard and mouse inputs. Currently, it has limited functionality with four apps: Spotify, Uber, DoorDash, and Midjourney.

  • What is the price of the Rabbit R1 and what additional costs might a user incur?

    -The Rabbit R1 costs $200 with no subscription fee. However, users need a separate SIM card for cellular functionality, which incurs a monthly data cost.

  • What is the 'Teach Mode' of the Rabbit R1?

    -Teach Mode is a feature that allows the Rabbit R1 to learn from a user's actions on their computer, so it can later repeat those actions. However, this feature is not available at the time of the review.

  • What is the significance of the Rabbit R1's bright orange color?

    -The bright orange color is a design choice by Teenage Engineering, which is known for its quirky and friendly looking devices. The color is so vibrant that it is difficult to accurately represent in photos.

  • How does the Rabbit R1's screen function?

    -The Rabbit R1 features a 2.9-inch diagonal TFT screen that displays the text of answers, a bouncing rabbit animation to indicate it's awake, and allows for touch typing in terminal mode.

  • What is the reviewer's stance on buying tech products based on their future potential?

    -null

  • What is the reviewer's opinion on the trend of companies delivering unfinished products?

    -The reviewer is critical of this trend, stating that it makes products difficult to review and can be frustrating for consumers who pay full price for a product that is not fully functional at the time of purchase.

  • What is the comparison made between Rabbit R1 and Tesla's initial approach to autopilot technology?

    -The reviewer compares Rabbit R1's strategy of getting devices into the hands of consumers to Tesla's approach, where they sold cars that were popular for other features and used the autopilot as a beta test to collect data and improve their systems.

Outlines

00:00

🤖 Introduction to the Rabbit R1 AI Device

The Rabbit R1 is a portable AI assistant device that shares similarities with the Humane Ai Pin but also has distinct features. It is a lightweight plastic cube designed by Teenage Engineering, with a recognizable design and a single button for interaction. The device has a built-in screen and a SIM card tray, and it sends requests to the Cloud, providing quicker responses than the Humane Ai Pin. It also includes a scroll wheel for UI navigation and a swiveling camera for multimodal interactions, such as identifying objects or summarizing text from a screen.

05:00

🔋 Critique of the Rabbit R1's Battery and Features

Despite its innovative features, the Rabbit R1 has significant drawbacks, including poor battery life and a lack of basic functionalities. The device struggles with battery drainage, requiring frequent charging and often being dead by morning. It also lacks the ability to set alarms or timers, record videos or photos, send emails, or include a calendar. The AI can sometimes provide incorrect answers, which is a common issue with this category of devices. The design, while quirky and appealing, features an analog scroll wheel that is frustrating to use, and the device lacks a back button, requiring users to scroll back to the top each time they want to navigate back.

10:01

📱 Touchscreen Limitations and the Concept of a Large Action Model

The Rabbit R1 features a touchscreen that is underutilized, restricted primarily to typing in terminal mode. The device's user interface could be significantly improved with touchscreen functionality, but currently, it is limited. The Rabbit R1 is also introduced with the concept of a 'large action model,' which theoretically allows the AI to perform actions within apps as a human would. However, with limited training data, the device currently only supports integration with four apps: Spotify, Uber, DoorDash, and Midjourney. The company has plans for future expansion and a 'Teach Mode' that would allow the AI to learn from user demonstrations, but these features are not yet available.

15:02

🚧 The Issue with Half-Baked Tech Products

The review discusses the trend of tech companies releasing products that are not fully functional upon launch, which makes them difficult to assess and review. The Rabbit R1, while promising in its concept, is criticized for its incomplete state at the time of release. The reviewer expresses frustration with this trend, where consumers pay full price for products that are not yet fully developed. The Rabbit R1 is positioned as a more affordable option compared to other AI assistants, which might make it an easier gamble for consumers to take. However, the reviewer advises buying a product based on its current capabilities, not its promised future potential.

Mindmap

Keywords

💡AI in a box

The term 'AI in a box' refers to a portable device that contains artificial intelligence capabilities, allowing it to perform tasks and answer questions without being connected to a larger system. In the video, the Rabbit R1 is described as an 'AI in a box,' emphasizing its standalone functionality and portability.

💡Virtual Assistant

A 'virtual assistant' is an AI-powered software agent that can perform tasks or services for users, such as answering questions, setting reminders, or providing information. The Rabbit R1 is characterized as a virtual assistant in a box, highlighting its ability to interact with users through voice commands and text responses.

💡Multimodal AI

Multimodal AI refers to artificial intelligence systems that can process and understand information from multiple sensory inputs, such as text, voice, and visuals. The Rabbit R1 is described as multimodal because it can answer questions based on text, voice, and visual inputs, like recognizing a plant from an image.

💡Scroll Wheel

A 'scroll wheel' is a user interface device that allows for the navigation of digital content by scrolling up or down. In the context of the video, the Rabbit R1 features a scroll wheel for navigating through its user interface, which is a key part of its unique design.

💡Swiveling Camera

A 'swiveling camera' is a type of camera that can rotate or pivot, allowing for a wider range of viewing angles. The Rabbit R1 has a swiveling camera that enables the AI to visually analyze objects or scenes when answering questions, adding to its multimodal capabilities.

💡Large Action Model

The 'large action model' is a concept where an AI system is trained to not only understand language but also to perform actions based on that understanding. The Rabbit R1 is said to have a large action model, which is intended to allow it to interact with apps and services on behalf of the user, although this feature is still in development.

💡Battery Life

Battery life refers to the amount of time a device can operate before needing to be recharged. The video script mentions the Rabbit R1's battery life as a significant drawback, highlighting the need for frequent charging and the inconvenience it poses to users.

💡SIM Card

A 'SIM card' is a small removable card that contains a mobile identity and subscription information, allowing for cellular network access. The Rabbit R1 requires a separate SIM card for cellular connectivity, which is an additional cost to consider, and it is a key component for the device's operation outside of Wi-Fi environments.

💡Touch Screen

A 'touch screen' is a display that allows users to interact with a device by touching the screen with their fingers. The Rabbit R1 has a touch screen, but its functionality is limited primarily to typing in terminal mode, which is a point of criticism in the video for its underutilization.

💡Teach Mode

Teach Mode is a proposed feature for the Rabbit R1 where the device would learn from observing a user's actions on a computer, then replicate those actions later. It represents an advanced level of personalization and AI learning, but as mentioned in the video, it is not yet available.

💡Product Roadmap

A 'product roadmap' is a strategic plan that outlines the future development and release of a product's features. The video discusses the Rabbit R1's product roadmap, which includes features like a generative UI and Teach Mode, indicating the company's vision for the device's future capabilities.

Highlights

The Rabbit R1 is a portable AI assistant device designed by Teenage Engineering, featuring a recognizable and quirky design.

The device is about the size of a stack of Post-its, lightweight, and made of plastic with a single button for interaction.

It has a built-in screen for displaying answers and a SIM card tray alongside a USB Type-C port for charging.

The Rabbit R1 is faster in answering questions compared to the Humane Ai Pin, although starting from a low baseline.

The device can connect headphones via Bluetooth and has a built-in speaker, showing text answers on the screen.

A bouncing rabbit on the screen indicates that the device is awake, and a button press puts it to sleep.

Settings are accessed by shaking the device like an Etch A Sketch, utilizing a scroll wheel and button for navigation.

The device includes a scroll wheel for UI navigation and a swiveling camera for multimodal interaction.

The AI can use vision to answer questions about objects it sees, such as identifying a Monstera deliciosa plant from an image.

The Rabbit R1 can summarize text from a computer screen or articles when asked, showcasing its text-to-speech capabilities.

Despite its innovative features, the device suffers from poor battery life, requiring multiple charges per day.

Lacking basic features like alarms, timers, video/photo recording, email, and calendar functionality.

The device is prone to providing incorrect answers confidently, a common issue with AI assistants.

The scroll wheel is frustrating to use due to its slow response and lack of haptic feedback.

The touchscreen is underutilized, only allowing keyboard input in terminal mode, despite its potential for improved UI interaction.

The Rabbit R1 is priced at $200 with no subscription fees, a significant drop from the Humane Pin's cost.

The device is made of plastic with a basic camera and speaker, and uses a low-end MediaTek chip, reflecting its price point.

The Rabbit R1 introduces a 'large action model' concept, aiming to perform tasks within apps as a human would.

Currently, the device supports limited apps like Spotify, Uber, DoorDash, and Midjourney, with more in development.

The 'Teach Mode' feature, which allows the device to learn from user actions, is not yet available.

The reviewer criticizes the trend of releasing unfinished tech products and the challenge it poses to both consumers and reviewers.

The Rabbit R1 represents a gamble on future potential, with the current functionality being limited but offering room for improvement.

The advice given is to purchase tech products based on their current capabilities, not on promised future features.

Transcripts

00:00

(upbeat music)

00:03

- So, this is the Rabbit R1.

00:06

And it's another AI in a box, and it's a sign of the times.

00:11

So this thing here,

00:11

it has a lot in common with the Humane Ai Pin,

00:15

for better or for worse.

00:16

But it's also supposed to have two specific things

00:19

that are supposed to set it apart from that.

00:23

Supposed to, anyway.

00:25

And we have to talk about it.

00:26

(letters clicking) (tense music)

00:29

So, stop me if this sounds familiar at any point,

00:31

but check this out.

00:33

This is a virtual assistant in a box.

00:37

But it's a different box.

00:38

This one is not wearable.

00:39

It's a thing you have to carry around with you

00:41

in your pocket, like a smartphone.

00:43

I'd say it's about the size of a stack of Post-its, maybe.

00:46

But it's this lightweight plastic cube,

00:48

designed by Teenage Engineering.

00:50

And boy, do people love some Teenage Engineering right now.

00:53

And it's super...

00:54

I mean, you can tell exactly what it is from a mile away.

00:57

This thing is very recognizable.

00:58

The thing has one button on the right side.

01:00

That's the one you press and hold to ask it questions.

01:03

And then instead of a projector,

01:04

it actually does have a built-in screen here.

01:07

And instead of built-in cellular,

01:08

it has a SIM card tray right alongside

01:10

that USB Type-C port for charging.

01:12

I'd say it does seem to send basically

01:14

every single request to the Cloud, but I will say,

01:16

it does feel significantly quicker

01:18

to answer questions than the Humane Ai Pin.

01:21

Now, that is a very, very low bar to clear,

01:25

but it's consistent, it's consistently quicker.

01:28

How far away is the moon?

01:34

- [AI Voice] The average distance

01:35

between Earth and the Moon

01:36

is approximately 384,400 kilometers.

01:40

- You can connect headphones via Bluetooth,

01:42

or turn the built-in speaker all the way up, like I have.

01:45

But it also shows you the text of your answer

01:47

on the screen, right underneath this bouncing rabbit,

01:50

and the time in the battery.

01:51

Matter of fact, it basically always shows

01:54

this bouncing rabbit all the time.

01:56

It's how you know it's awake.

01:57

You hit that button once to put it to sleep.

02:00

Or it would do that automatically after a few seconds.

02:02

Then hit the button again to wake it up.

02:04

Boom, just the rabbit waiting for you there.

02:06

And then to get to settings,

02:07

there's no button to press for that.

02:09

There's no gesture on the screen or anything.

02:11

You just shake it like an Etch A Sketch.

02:14

And as long as you do that,

02:15

then from there you can scroll up and down

02:17

with this scroll wheel, and select with the button.

02:19

That's how you get around the UI.

02:22

So, the other two things it has

02:23

are the scroll wheel and a swiveling camera.

02:26

So the scroll wheel for navigating around the UI

02:28

instead of the screen is...

02:31

I'll get to why it's odd in a second.

02:32

But then of course, the AI assistant is multimodal,

02:36

and so you can use vision to answer questions

02:39

about what it sees.

02:40

So we've seen this before.

02:41

With this one, you double-tap to swivel it open,

02:44

and then press and hold to ask.

02:46

I don't know. What type of plant is this?

02:50

- [AI Voice] Taking a look now.

02:53

The plant in the image appears to be a Monstera deliciosa,

02:56

also known as a Swiss cheese plant.

02:58

This is a type of tropical climbing evergreen plant

03:02

native to southern Mexico and parts of Central America.

03:05

The large glossy green leaves with distinctive-

03:07

- As a reviewer testing this thing,

03:09

I just feel like that DJ Khaled clip after a while,

03:11

where I'm just pointing at stuff,

03:12

like, "Okay, what is this?

03:14

And what is this?"

03:16

- And what is this?

03:16

- [Speaker] Those are pickled banana peppers.

03:18

- [DJ Khaled] And what is this?

03:19

- [Speaker] Berries and seeds.

03:21

- [DJ Khaled] And perhaps what is this?

03:22

- [Speaker] Water.

03:23

- But I've also, I've just pointed it at my computer screen

03:25

with a really long email on it,

03:27

and asked it for a summary,

03:28

and it just reads an instantly and gives me a summary of it.

03:31

It's done that with articles too.

03:32

I think that's pretty cool.

03:34

But yeah, it's just a...

03:35

It's an AI in a box.

03:37

It's best at just answering questions.

03:40

(text clicking) (soft music)

03:42

So yeah, this thing is also bad at a lot of stuff.

03:45

And this list is gonna sound pretty familiar.

03:47

Somehow, the battery life is just as bad as the Humane Pin.

03:51

It has a thousand million power battery inside

03:54

and it's brutally bad.

03:55

Like it's already bad enough

03:56

when you're carrying around a whole nother device

03:58

alongside your smartphone,

04:00

but when this one can sit in front of you doing nothing,

04:03

and the battery just visibly is draining

04:06

and dies in like four hours...

04:08

Then you have to charge it multiple times per day,

04:10

and it's still dead when you wake up in the morning.

04:12

Like, it's just exhausting.

04:13

Also, it took 45 minutes to charge this tiny battery

04:17

from zero to a hundred.

04:18

And it's also just straight up missing

04:20

a ton of what I would consider just basic features.

04:23

Like it can't set alarms, can't set timers,

04:25

it can't record videos, can't record photos,

04:28

can't send emails, there's no calendar built in.

04:31

There's just a lot of things that I would want

04:33

an assistant to do...

04:34

Not here.

04:35

And of course, being an AI assistant,

04:37

it also does still hallucinate,

04:39

and confidently answer questions wrong.

04:42

Like my baseline of asking a question

04:44

that I know the answer to,

04:45

and then getting the wrong answer happens all the time.

04:49

Which is one of the downfalls of this category.

04:50

So this device was designed by Teenage Engineering.

04:54

And they're really leaning into that.

04:55

Like that's why...

04:56

I mean, it's this bright orange, it is really quirky,

05:00

friendly looking device, I would say, intentionally.

05:04

But yeah, they really love their analog controls.

05:06

And this scroll wheel here doesn't surprise me,

05:10

but it's actually really frustrating to use.

05:13

First of all, it protrudes out the back a little bit,

05:16

as you can see.

05:16

Which looks kinda cool.

05:18

But that means if it was really sensitive,

05:20

that it would actually scroll if you just put it down

05:22

on a table or something.

05:24

So they've dialed down the sensitivity of the scroll wheel,

05:26

so it's actually super slow.

05:29

Like there's a surprising amount of scroll motion

05:31

required to go down one line in the settings.

05:34

And then there's no haptic feedback here as you're scrolling

05:36

to help you feel it out.

05:38

So fine, okay, you can get used to that.

05:40

You use the button to select...

05:42

But then you've also noticed,

05:43

there's no back button anywhere on this device.

05:46

So to go back up a level,

05:47

you have to scroll all the way back up to the top

05:50

every single time.

05:53

That also gets annoying.

05:54

And then to change brightness or volume,

05:57

you actually need two hands.

05:58

So you just go into brightness to select,

06:00

and you hold the button with one hand,

06:03

and then scroll the scroll wheel

06:05

with the other hand to adjust brightness.

06:07

Which, yeah, you can't do it with one hand.

06:10

It works, so you can learn it, and you can say, you know,

06:13

it's this quirky UI it's got going on.

06:15

But I feel like a lot of these problems would be solved

06:18

if this was a touch screen.

06:21

So what if I told you this is a touch screen,

06:24

and they just don't really let you use it

06:26

for almost anything?

06:27

Like moving through these menus would be easier

06:29

if I could just tap what I wanted to, right?

06:31

Going back after a long scroll would be easier

06:34

if I could just flip scroll back to the top of a list,

06:36

and hit back.

06:37

But you can't.

06:38

The only thing you can use the touchscreen for

06:40

is basically typing on the keyboard in terminal mode.

06:43

So with terminal enabled, you can turn it sideways,

06:47

a keyboard pops up, and you can type your questions

06:49

and get text answers.

06:51

Which is great.

06:52

And you can move between letters with the scroll wheel.

06:54

It's kind of a neat feature, I guess.

06:56

But why can't we use the touch screen for anything else?

07:00

Is it just them trying not to look too much

07:02

like a smartphone?

07:04

Maybe.

07:06

Probably.

07:07

(text clicking) (tense music)

07:09

So, okay, what's the point?

07:12

Why does this thing exist?

07:12

If it's so similar to the other one

07:14

that was also so bad, what are we doing here?

07:17

And really, there's two things

07:18

that I think they're hoping will be the things

07:20

that set it apart from the other one.

07:23

That should, in the words of their co-founder, Jesse...

07:26

- But I don't think MKBHD will say

07:29

this is the worst device he ever reviewed so far.

07:31

(people laughing)

07:32

- Which, low bar, but okay.

07:33

Those two things are the price tag,

07:37

and the large action model.

07:38

So the Humane Pin was so easy for everyone to dunk on

07:42

because it costs as much as a phone.

07:44

It was $700 with a $24 a month subscription

07:48

to not turn into a brick.

07:50

It's just insane.

07:51

So, this one right here costs $200 and no subscription.

07:57

So, you know, okay, that hits a little different.

08:00

But also, you can tell it's $200.

08:03

First of all, you will need a separate SIM card

08:05

to get it to work on cellular.

08:07

So while there is no subscription fee for Rabbit

08:09

to keep the device working,

08:11

that is still a fee you're gonna pay every month

08:13

to get data outside of wifi.

08:15

But then, okay, the unboxing experience

08:17

is extremely minimal.

08:18

It comes in a single cardboard box

08:20

with a cassette tape looking plastic container,

08:22

that doubles as a stand,

08:24

but there's literally nothing else.

08:25

No charging brick, no USB-C cable, no stickers,

08:29

no paper instruction manual, nothing at all.

08:31

And then the R1 itself is made of plastic.

08:34

Not to say that it's not built well.

08:36

You know, there's no flexing or creaking

08:37

or anything like that.

08:38

But it's definitely plastic.

08:40

The camera, very basic.

08:42

The speaker, very cheap.

08:44

You know, it's a low end MediaTek chip inside.

08:46

The same one that's in the Moto G8 Power Lite,

08:49

which is like $150 phone.

08:51

Hardly any battery, as we learned.

08:53

And no fancy fast charging or wireless charging.

08:56

And it comes in one color, really bright orange.

08:59

Like this ridiculously saturated bright orange.

09:03

I'm not even joking, the color you're seeing

09:05

in this YouTube video on your screen,

09:07

it's gonna be the best I can do

09:09

with my own color correction.

09:10

But it's legitimately hard to photograph.

09:13

Like the phone camera doesn't really

09:16

turn saturation up high enough

09:17

to really represent it accurately in real life.

09:19

It's the brightest orange thing

09:21

you've ever seen in your life, I guarantee it.

09:23

Now, if you want something a little more low key

09:25

than the neon orange,

09:26

channel sponsor dbrand actually does have you covered here.

09:28

I've got this black camo looking one here,

09:31

and now the orange bits that show through

09:32

are kind of like more accent pieces, which is pretty dope.

09:35

But fun fact, dbrand also collaborated with Rabbit

09:37

ahead of the release to make sure

09:38

screen protectors for this thing

09:40

were also available on launch day.

09:42

So, that includes mine.

09:43

If you wanna check it out,

09:44

screen protectors or skins for the R1,

09:47

you can hit up the link below.

09:48

But yes, speaking of the screen,

09:49

this is a 2.9 inch diagonal TFT screen,

09:54

with no auto brightness.

09:55

So, yeah, I think you get the idea.

09:58

(letters clicking) (soft music)

10:00

So you know how the Humane Ai Pin had no apps?

10:04

Well, this thing also has no apps.

10:06

But the other thing that they're hoping

10:08

will separate it from the other stuff,

10:10

is what they're calling a large action model.

10:12

So you know, large language models are just the words,

10:16

the language that we use.

10:17

So it's an AI that can take our natural language,

10:19

and process it, and turn it into words in return.

10:23

And that's how we interact with it.

10:24

So a large action model is supposed to take our words,

10:29

and then process it, and turn it into actions.

10:32

That's the theory.

10:33

So it's basically going to be able to use apps for you

10:37

like a human would,

10:38

based on what you tell it to do.

10:39

Now, importantly, this isn't the same as an API,

10:43

where you know, a company might work with the others

10:45

to have some sort of a plug into their services

10:48

so that they work together.

10:49

Because an API can be a little too restrictive,

10:52

or not give all the features, or just not quite...

10:54

I mean, there's lots of good reasons

10:56

why they don't wanna rely on an API.

10:58

So what this is supposed to be doing,

11:00

is just going in and using the app, just like a human,

11:04

essentially with like a mouse and a keyboard.

11:06

Think of it as like a virtual agent.

11:08

Honestly, I think it's a really cool idea.

11:10

I think it's a really, really cool idea.

11:13

You know, large language models have been trained,

11:15

with all their data,

11:16

to talk back to us just like humans would,

11:19

and some are very convincing.

11:21

So a large action model, in theory,

11:23

should be able to use these apps and services

11:27

just like a human would.

11:29

From Spotify to Twitter, to your banking app,

11:33

to whatever else.

11:34

It's already good enough at recognizing major UI elements

11:39

like a play button or a buy button and things like that.

11:42

So with enough training data, could get really good.

11:47

With enough training data, could get really good.

11:51

But the thing is,

11:52

they don't have a ton of training data yet.

11:55

So as of right now, they have made four apps available.

12:01

Four.

12:02

They have this online portal called the Rabbit Hole,

12:04

where you can log into and enable all of them.

12:07

Spotify, Uber, DoorDash and Midjourney.

12:12

So the Rabbit can talk to these four apps,

12:14

and has a UI on the screen for doing so

12:17

in the exact way that we're theorizing.

12:19

So it'll play Spotify songs.

12:21

Like I can ask it for a song,

12:23

and it'll try to play the right one,

12:24

and it'll show you any info you need on the touch screen

12:28

to confirm or refine the actions you're taking.

12:30

But even now, they kinda work.

12:32

Like, I've already had issues with it

12:34

playing the wrong song.

12:35

I've also seen other people already have issues

12:37

with the DoorDash app getting things wrong,

12:39

which is even more frustrating.

12:41

I can't imagine how frustrating the Uber app

12:43

messing up would be.

12:44

It definitely seems like it needs more training data.

12:46

And that's just for these four starting apps.

12:48

Now, Rabbit says they have

12:51

800 different apps trained already,

12:54

but they just haven't built a UI for it yet,

12:57

so it's not available on the Rabbit yet.

12:59

And so that's still in the works.

13:02

Now they've also theoretically

13:03

started working on something called a generative UI,

13:06

where essentially it can recognize

13:08

what type of app it is,

13:10

and then build a UI for it so Rabbit doesn't have to...

13:13

But that's also still theoretical,

13:15

and also on the roadmap and still in the works.

13:18

And then if there's another app or service

13:20

that you want your Rabbit to be able to do,

13:22

like something for work,

13:23

or something super niche that you've invented,

13:26

they've talked about something called "Teach Mode".

13:29

Which is basically having the Rabbit watch you

13:33

do what you want it to do on your mouse and keyboard,

13:36

and then it learns from what you did,

13:38

and can then repeat your actions later, which is sick.

13:41

Unfortunately, you can probably already

13:43

see where this is going,

13:44

Teach Mode is also not available as of right now.

13:47

This is something they're still working on

13:49

vetting and finishing.

13:49

And so that's just later this year, also in the works.

13:53

So as of right now,

13:55

this device has none of that stuff,

13:59

just the four kind of half working apps

14:01

that we went over at the beginning.

14:04

What are we doing here?

14:06

(letters clicking) (soft music)

14:08

Okay, I'm gonna try not to turn this into a rant,

14:11

but I feel like we need to acknowledge at least,

14:14

that a lot of these tech companies

14:17

are developing tech kinda backwards.

14:21

Like they're delivering such unfinished products,

14:24

that it actually makes them nearly impossible to review.

14:27

Like, it feels like it used to just be,

14:29

make the thing, and then put it on sale.

14:32

Now it's like put it on sale,

14:34

and then deliver the half-baked thing,

14:36

and then iterate and make it better,

14:38

and hopefully with enough updates, then it's ready,

14:40

and it's what we promised

14:41

way back when we first started selling it.

14:43

And then this whole period in the middle is a mess.

14:46

And it's across all kinds of product categories too.

14:48

We've seen this with gaming.

14:49

Like huge studios are delivering half-baked games saying,

14:53

"Oh, you know, it's an alpha version.

14:55

There will be updates."

14:57

But meanwhile, it's a full price AAA game

14:59

that's just gotten an unacceptable number

15:01

of bugs and issues.

15:03

It's also happening with cars.

15:05

And vehicles getting announced,

15:06

and then delivering with like a half-finished state,

15:09

where you just don't get a lot of the features

15:11

that you paid for.

15:12

And they're eventually coming soon,

15:14

with a software update.

15:15

You know, smartphones, obviously,

15:16

we've been seeing this for years,

15:18

but it does seem like now more than ever,

15:20

there's at least one feature,

15:23

one major feature of every smartphone launch

15:25

that gets announced,

15:26

but that's not coming until later in the year.

15:28

And now these AI based products are at like the apex

15:32

of this horrible trend where,

15:34

the thing that you get at the beginning

15:36

is like borderline nonfunctional

15:39

compared to all the promises and all the features

15:42

and all the things that it's supposed to maybe someday be.

15:45

But you still pay full price at the beginning,

15:47

which is what makes it so crazy.

15:48

And that sucks not only obviously

15:50

for the people buying it at full price at the beginning,

15:52

but also for reviewing these things in general.

15:55

Like, how do you assess a product

15:58

where the version of what's promised in like three years,

16:03

what it could be is amazing,

16:05

but the version that's being delivered now is dog water.

16:09

Like, how do you connect those dots?

16:12

Do you even connect those dots?

16:13

Are you supposed to give them the benefit of the doubt?

16:16

I don't know. What are we even doing here?

16:17

I mean, I guess on one hand,

16:18

it is good that tech products,

16:21

these things can get better over time and improve,

16:23

and the thing you buy can be better tomorrow.

16:26

That's all great.

16:27

That didn't used to be true,

16:28

and that's beautiful and everything.

16:29

But on the other hand,

16:31

it's just, the other side of that coin,

16:33

is some very unfinished products being delivered,

16:37

and I just feel like it's gonna get worse

16:40

before it gets better.

16:41

(letters clicking) (tense music)

16:43

So, let me just say,

16:45

I am personally very excited

16:48

by the idea of a super personalized AI assistant

16:53

that can do everything that a human assistant could.

16:56

That's my dream. I want that to happen.

16:58

And I'm interested that these different companies

17:00

are approaching it from different angles.

17:03

And hopefully, we'll get there eventually.

17:05

But this is also so clearly going to take

17:07

a lot of time and effort and technical development,

17:11

and also a lot of data.

17:13

Like, I've said this already in the Humane review,

17:16

I'm pretty sure, which is a good assistant needs to know

17:19

everything about you, whether it's human or virtual.

17:22

Like it needs to know your preferences,

17:24

and your location,

17:25

and what you're doing, and what you like to do.

17:27

And every single little thing.

17:28

It's so much data.

17:29

So as for the Rabbit...

17:32

I think that's why they focused on getting the price so low.

17:35

Like, this stuff is all a tough sell,

17:36

because you're taking a gamble

17:37

on what the product could be someday.

17:39

But for the Pin, you have to drop $700, and $24 every month,

17:45

on the chance that it might someday get there.

17:48

Like that's a lot of money.

17:49

But for this cute little thing, 200 bucks.

17:52

Feels like a much easier investment to make.

17:55

A much easier pill to swallow,

17:56

to just take a gamble on it.

17:58

And on the off chance it turns out to be awesome

18:00

in a couple of years, it'll feel like it's worth it.

18:02

You know, on the chance that some of those 800 apps

18:05

are super useful.

18:07

Or you know, even though you already have a smartphone,

18:10

maybe the Teach Mode lets it do something

18:12

with a single button press

18:13

that you could never dream of before,

18:15

then it could be totally worth it.

18:16

The whole situation reminds me kind of...

18:18

If you look back at the beginning

18:20

of when Tesla started shipping cars

18:23

with all the autopilot stuff.

18:26

They were selling cars,

18:27

and they got popular for other reasons.

18:28

It was a good electric car with good range,

18:30

and a good charging network.

18:32

And then people started sort of beta testing

18:34

this autopilot stuff,

18:36

and Tesla starts collecting

18:38

millions of miles and hours of data,

18:41

from how it works in the real world.

18:42

And so all of that data helped their systems learn,

18:45

and get the head start that they need,

18:47

on trying to make the best autopilot software.

18:49

So the Rabbit, I'm sure,

18:51

would like to use their price advantage,

18:52

and get as many of them out

18:54

into the real world as possible,

18:55

into people's hands,

18:56

so they can start beta testing,

18:57

and having it trained and getting good

18:59

at a variety of tasks.

19:01

But the Rabbit's problem is,

19:03

it doesn't have like the other reason

19:05

why people would get it anyway today.

19:08

So it's like a kind of chicken and egg problem.

19:11

So if you want my advice, you've heard it before.

19:15

Buy the product based on what it is today,

19:19

and not what it's promised to be in the future.

19:22

And with this category, that's like the hardest thing to do,

19:26

the hardest thing to keep in mind.

19:27

'Cause the promise is so big.

19:29

But that's still my advice.

19:31

And I guess we still also have yet to see

19:33

what big companies like Google and Apple

19:37

are gonna try to do in this space, probably this year.

19:39

So, with all of that, we'll see.

19:44

Thanks for watching.

19:45

Catch you in the next one.

19:47

Peace.

19:48

(mellow music fades)

Rate This

5.0 / 5 (0 votes)

Etiquetas relacionadas
AI AssistantProduct ReviewTech InnovationRabbit R1Smart DevicesVoice AssistantApp IntegrationUser InterfaceBattery LifeConsumer TechGadget ReviewAI LimitationsProduct Development
¿Necesitas un resumen en español?