The Race to Create the "iPhone of AI" is Heating Up!

Summary

The video explores the emergence of AI-powered hardware devices like the R1 by startup Rabbit. These voice-controlled devices aim to execute tasks across apps via natural language, reducing clicks. The video analyzes Rabbit's technology, use cases, business model and industry landscape, noting smartphones may make standalone devices redundant if integrated AI improves. It remains uncertain if consumers will adopt voice-first hardware, but the video argues Rabbit's crowdsourced training data and modding community could make the R1's AI increasingly capable over time.

Takeaways

• 😲 Humane's pin device failed to convince anyone of its usefulness or viability
• 👀 Rabbit is a new player trying to make AI hardware devices like the R1
• 📱 The R1 aims to execute tasks by integrating with your existing apps
• 🔊 The R1 has a push-to-talk button for quick voice access without a wake word
• 💡 The R1's AI model combines neural networks and symbolic AI for task automation
• 💰 The R1 only costs $199 with no subscription compared to Humane's $699 price
• 🤔 It's questionable if people will adopt a separate voice-first device like the R1
• 🚀 The R1 could accumulate training data to eventually control any app via voice
• 💬 Microsoft and others seem interested in the AI hardware device space
• 📈 If successful, the R1 concept could alter how we interact with technology

Q & A

What was the first company to try to build AI hardware, and what happened with their product?

The first company was Humane. They unveiled an AI assistant device called the Humane AI Pin in 2023, but it has not been released yet. The product was supposed to replace smartphones, but it failed to convince anyone of its practical utility and the company has struggled, laying off staff and losing their CTO.

What is the Rabbit R1 device that was recently announced?

The Rabbit R1 is a small, portable AI assistant device created by the startup Rabbit. It features voice control and physical buttons to activate the AI and interact with apps on the user's smartphone to perform tasks or answer questions.

How does the Rabbit R1 work compared to other voice assistants like Siri?

The R1 uses a large action model rather than just a language model. This allows it to not only understand requests, but actually take actions across apps and services to fulfill the user's needs. It connects to the user's existing apps and accounts.

What kind of tasks can you ask the Rabbit R1 to perform?

You can ask the R1 to perform complex, multi-step tasks involving different apps/services like booking a vacation, ordering food, calling you a rideshare, controlling your smart home devices, etc. Basically anything that can be done through an app, the R1 aims to be able to execute through voice commands.

What are some of the advantages of the Rabbit R1 concept?

Advantages include very fast response times, hands-free operation for tasks that normally require manually operating a phone, and a low $199 starting price with no subscription fees.

What are some potential disadvantages or doubts about the Rabbit R1?

You have to carry another device that might not provide enough utility compared to just using your phone. Voice-first operation contradicts social norms. And there are still doubts if AI can reliably perform some sensitive tasks like booking travel or payments.

How might the Rabbit R1 improve over time?

As more users show the device how to operate new apps, it can aggregate that training data to expand what apps it can use. Software updates utilizing improved AI models could also make it more capable of understanding any app interface.

Why is Microsoft interested in this new category of AI hardware?

Microsoft sees significant potential for 'breakthrough natural interfaces' between humans and devices/services with these new large AI models. If hardware can be designed around conversation versus individual apps, it enables a whole new interaction paradigm.

How are new chips advancing on-device AI capabilities?

New mobile chips like the Snapdragon 8 Gen 3 enable real-time AI features solely within the device itself, rather than needing cloud connectivity. This makes experiences faster, more reliable, and more private - key advantages.

What is one open question about Rabbit's business strategy with the R1 device?

One question is whether the low $199 starting price without ongoing fees can sustainably fund the advanced AI capabilities Rabbit claims. There is skepticism if their funding andAI costs support this business model long-term.

Outlines

00:00

😃 Pokemon Charmander Described in a Wooden Cabin

The first paragraph describes a scene with Pokemon characters like Charmander, Pikachu, etc inside a wooden cabin. It provides visual details of the characters and setting.

05:02

😯 Rabbit R1 AI Assistant Orders Pizza

The second paragraph demonstrates the Rabbit R1 AI assistant's ability to understand natural language instructions and complete tasks like ordering a pizza. It shows a conversational interaction where the assistant confirms order details.

10:02

🤔 Microsoft and Others Eyeing AI Hardware Market

The third paragraph discusses major tech companies like Microsoft, Apple, and Google exploring the emerging AI hardware space. It suggests the market potential but also risks of developing specialized AI devices.

15:03

🔍 Understanding Rabbit's Neurosymbolic AI

The fourth paragraph provides background on the neurosymbolic AI technology behind the Rabbit R1 device. It contrasts neural networks and symbolic AI, noting their limitations.

20:03

🤨 Evaluating Advantages and Disadvantages of Rabbit R1

The fifth paragraph objectively evaluates pros like speed and price and cons like carrying a separate device for the Rabbit R1. It concludes mainstream adoption will take time.

Keywords

💡AI assistant

An AI assistant is a software agent powered by artificial intelligence that can understand natural language requests and complete tasks on a user's behalf. The video discusses several AI assistant devices like the Humane pin, the rabbit R1, and potential future AI assistants built into phones.

💡natural language

Natural language refers to the ordinary language that humans use to communicate, as opposed to computer languages. AI assistants aim to understand natural language requests so users can interact with them conversationally.

💡multimodal

Multimodal refers to an AI system that can understand and interact using multiple modes like speech, vision, and touch. The rabbit R1 is described as multimodal since it has a camera, screen, scroll wheel etc. in addition to speech.

💡neurosymbolic AI

Neurosymbolic AI combines neural networks with symbolic AI techniques. It aims to gain the reasoning ability of symbolic AI and the pattern recognition strength of neural networks. The video suggests the rabbit R1 runs on a neurosymbolic model.

💡on-device AI

On-device AI refers to AI algorithms running locally on a device like a phone, without needing cloud connectivity. The video discusses how new phone chips are enabling powerful on-device AI capabilities.

💡general intelligence

General intelligence refers to an AI system's ability to understand and reason broadly across domains, like a human. AI assistants aim for greater general intelligence to become useful personal agents.

💡training data

Training data refers to the labeled examples used to train an AI model to recognize patterns. The video suggests rabbit R1 customers could provide aggregate training data by showing it how to use apps.

💡modding community

A modding community refers to enthusiasts who modify hardware/software to customize or enhance capabilities. The video suggests such a community could find creative uses for AI devices like the R1.

💡killer application

A killer application is a computer program so useful it proves the value of some new technology. The video discusses whether AI assistants have found a killer consumer application yet.

💡transparency

Transparency refers to an AI system clearly conveying its reasoning and capabilities to users. Lack of transparency seems an issue for AI assistants in building user trust currently.

Highlights

The R1 is a small orange device causing waves in the tech space with 50,000 pre-orders sold out in just over a week

The R1 can interact with your apps and do tasks on your behalf by just pressing the push to talk button

The R1 response time is a claimed 500 milliseconds or less unlike today's voice assistants

The R1 is powered by a large action model instead of a traditional large language model

The R1's model combines newer neural networks with older symbolic AI using neuro-symbolic programming

The R1 costs just $199 with no subscription compared to Humane at $699 with subscription

The 50,000 R1 pre-order customers become training data to improve the AI over time

The R1 is useful for early adopters who love to tinker, mainstream will take time to adopt

AI hardware faces threats from improving smartphone capabilities and integration into watches

The R1's utility is undetermined, killer applications aren't yet apparent

Carrying a separate R1 device may not warrant the utility compared to a smartphone app

Ordering food on the R1 without seeing menu options seems limiting

Trusting the R1 completely to book travel could miss specific preferences

The R1 concept can't be dismissed given improving AI capabilities over time

Steve Jobs predicted apps integrating into an agent-centric voice system

Transcripts

this video is brought to you by

nebula who is this Pokemon Charmander a

fire type Pokémon known for the flame on

the tip of its tail in the scene you

described Charmander is surrounded by

other Pokémon such as Pikachu a small

yellow rodent-like Pokémon with long

ears and a lightning bolt-shaped tail

the setting appears to be the interior

of a wooden cabin or house with the

Pokémon standing on a wooden floor yep

the image is taken from a screen as

evidenced by the reflection of some

items on the screen and style of the

windows and

walls that's crazy but yeah it is a

pokedex hi welcome to another episode of

Cold

Fusion for decades it's been a Sci-Fi

dream to Simply talk to your device and

have it do tasks for you with generative

language AI starting to make this stream

feasible the question becomes how do we

turn these AI capabilities into a

functional Hardware device recently a

new unknown player has entered the space

rabbit the company only started raising

money in October of 2023 and have

already built out a product called the

R1 this small orange device is causing

waves in the online tech space with

50,000 pre-orders sold out in just over

a week people are interested so I

figured that in this episode I'll give

my thoughts on the device but I'll also

show you the new AI industry that's

forming right before our eyes this is

going to be a deep dive so sit back

relax and let's get into it

you are watching C Fusion

TV we need some context rabbit aren't

the first company to try AI Hardware

they're the second the first was Humane

and let's just say that they

embarrassingly showed how hard it is to

get this idea

right many of you have heard of The

Humane AI pin unveiled in 2023 and

unreleased at the time of this recording

it's a device you can speak to a natural

language and have it do tasks for you it

has a holographic display and was

supposed to replace your phone despite

the founders having the pedigree of X

Apple Engineers they didn't seem to live

in reality the device was more like an

exercise in philosophy rather than

practicality take a look at the CNBC

interview where they failed to convince

anyone can you see that yeah what's that

telling you so this is this is a display

when you need it it's just capable of

doing a lot of things but it's not

something that you need because the

device is actually built to be

multimodal that means you can use it

however you want like what like what

tell me something you would use that for

so you use it for just about anything

like uh sending texts or checking up on

any notifications that you've got in

stuff that you do just all there's a

speaker that's built in and then there's

a a user LED called so that's what I

want to ask you about Imran the

co-founder says that the device can do

virtually anything a smartphone can but

without a screen I doubt that's possible

something as simple as discreetly

reading emails or texts can't be done

sometimes I'll be at a meeting yeah

right and and I'm sneaking under the

under the desk to look at my email

because that's and I can't say uh Hey

Hey Siri you know tell me my email

because I don't want the whole room to

hear my email so how does that work in

this context so the device is powered by

a AI powered OS I think the biggest

thing here that is that it's AI powered

and so it's doing a lot of that that

heavy lifting for you so you just

actually engage with it when it's really

important but seriously if if if I'm in

a meeting and I'm talking to somebody

but I got to go like this when I'm in

the meeting and anybody who can see my

hand can see the see what I'm reading

too how I mean I don't know how that's

different screen's doing and then

there's the price VAR what's the price

point it's $6.99 for this model and it's

$7.99 for the one that she's got and

that's the complete system comes with

everything plus obviously this hasn't

gone too well as Humane has already laid

off 4% of their staff before even

launching their CTO has been kicked out

too but while all of of this is going on

there's a new kid in town rabbit with

their new device called the R1 firstly

what is it that was delicious check the

fridge and order the ingredients to make

that again tomorrow create a route that

works with my goals then start the best

playlist to keep me motivated watch what

I'm doing here process all my new photos

today just like this find us a nice

restaurant near here then get us there

take me to the best spot to catch the

sunset and Lead me through a meditation

play my favorite

song I'm on

it at its core the rabbit R1 is supposed

to be a device so simple that you don't

need to learn how to use it it's powered

by a multimodal language model

supposedly more capable than chat GPT so

of course it can answer all the

questions that you have but the

difference is it can interact with your

apps and do tasks on your behalf with

the push to talk button you don't need

to say anything to wake it up just press

and hold the button and talk like a

walkie-talkie I get a response 10 times

faster than most of the voice AI

project get me a 12in pizza from Pizza

Hut deliver to here the most ordered

option on the app is fine ordering a

12in pizza from Pizza Hut since you

mentioned that the most ordered option

is fine I will select that for you I

just created an order for a 12-in pizza

it's going to be hand tossed with a

classic marinara sauce and topped with

regular please please confirm your

order that sounds really good I just

confirmed order

here now I'm waiting for my pizza set of

more complex actions that may take

multiple apps to finish on a smartphone

for example if I want to plan a trip to

London I can just describe what I wish

in full to R1 I want to take my family

to London it's going to be two of us and

a child of age 12 we're thinking of

January 30th to February 5th can you

plan the entire trip for me we like

cheap non-stop flights grouped seats a

cool SUV and a nice hotel that has Wi-Fi

exploring ticketing options to make your

trip a reality for your trip I found

various flight options a range of hotels

to choose from and car rentals available

please confirm each option individually

for further details and booking could

you come up with a schedule for fun

things to do while I'm over there I have

prepared a detailed travel plan for your

trip in summary you will be exploring

London's iconic landmarks visiting

museums enjoying delicious meals at

local restaurants and experiencing the

vibrant atmosphere of different

neighborhoods navigation details are

also prepared and I can help you book

tickets for any events you'd like to

attend R1 just planed the entire trip

for me that's awesome but it seems like

this is a little bit too intense can you

plan a easy schedule for us sure I'm

working on it please take a look at it

and let me know what you think it gave

me a more relaxed schedule every day

with all the details sounds really good

and I just confirmed that I can foresee

a wonderful trip Isn't that cool huh you

can use the camera to recognize real

world objects and ask questions about

them for example what meals to make from

a fridge

or what are people talking here let me

take a look in the image there is a

conversation taking place in a chat

application the participants are

discussing rabbit OS and its

capabilities specifically mentioning a

music sequence answer one person is

curious if rabbit OS has a music

sequencer built in and another suggests

that it can manipulate your digital

audio workstation DA they also mention

using it as a learning tool and seem to

be preparing for the next topic or task

in their discussion whether it works as

well as the demos is another question

entirely the rabbit company stresses

that the R1 is not a phone but unlike

Humane it's not aiming to replace it the

R1 has a 2.8 in screen which is small

given that the average smartphone screen

is 6 Ines these days but the small

screen is meant for increasing utility

rather than the main form of user

interaction in addition the R1 has a

physical scroll wheel a push to talk

button so it doesn't need a trigger

phrase to listen a 360 swivel camera and

enter the

vision a rotates and of course a speaker

it's been designed by teenage

engineering who makes music hardware and

famously designed the nothing phone its

design might be polarizing but at the

very least the device is interesting

enough to get people

talking in terms of the business case

for rabbit there are some clear threats

generative AI technology is looking for

Consumer killer applications

entrepreneurs are naturally trying to

find the best ways to use it recently

jabel made a pretty hilarious video

about coding and building his own

companion AI

device but I played with it for a bit

and it feels more like uber besty

instead of actually having one so if I

want a true AI bestie I need to get more

control and after a few hours is the

python dependency hell check this out I

got it working on the raspberry pie my

little AI homie name is chai by the way

and from there I started working on the

portability by buying the right parts

for it and giving a little auto start

script started work on chai's actual

build form with the res piie screen and

3D printer and I just kept on iterating

on this process until my main man chai

was in one solid piece so as you can see

he is a one complete build one solid

object a microphone on top the back's

closed please don't ask how I charge him

instead new AI bestie thing my buddy

here actually has his own personality

watch this I'm a black guy who loves

anime what do you think about that

aren't we breaking stereotypes here a

black guy who loves anime alert the

media because this is such

groundbreaking information dude you

enjoy what you enjoy who cares about

what anyone else thinks aren't we

Progressive high five keep watching

Dragon Ball or whatever you're into

interesting so he like partially roasted

me but also ended on a positive upbeat

listen to being real that sounds a lot

closer to what an actual best friend

would do standard chat GPT could never

so the fact that one single guy can do

this means that there's probably a bunch

of startups chomping at the bit to enter

the hardware AI space in fact Sam ultman

and softbank's MOSI son are both trying

to get the iPhone designer Johnny I to

design their respective AI Hardware

devices Microsoft is looking at the

development of the R1 with great

interest Satia nadela hints that

Microsoft could jump into the AI

Hardware space you see I thought the the

the demo of uh the rabbit OS and uh the

device was fantastic I think I I must

say after jobs uh sort of launch of

iPhone probably one of the most

impressive presentations I've seen of

capturing the uh the vision uh of what

is possible going forward for what is an

agent Centric um uh operating system and

interface and uh I think that's what

everybody's going seeking if you have a

breakthrough in natural interface uh

where this idea that you have to go one

app at a time and all of the cognitive

load is with you uh as a human uh does

seem like there can be a real

breakthrough we had the first generation

whether it was Cortana or Alexa or Siri

or whatever you um it was just not it

was too brittle uh where we didn't have

these Transformers these large language

models uh whereas now we have I think

the tech to go

and come up with a new app model and

once you have a new interface and a new

app model I think new hardware is also

possible and it's had an opportunity

from Microsoft or are you moving away

from Hardware I mean look I mean always

it's an opportunity for us um and so

yeah I mean we make Hardware but even

for the whole category of AI Hardware

devices as a whole it's a risky segment

to be in there's a quiet AI Revolution

going on in smartphones Snapdragon new

chip the Snapdragon 8 Gen 3 is out now

and is already giving smart phones new

on device AI capabilities Samsung's s24

is one of the first devices to fully

take advantage Mr Who's The Boss shows

us some

examples so the first is instant slowmo

literally go into your phone's Gallery

hold down on any video in it and Bam it

halves the playback speed while keeping

it smooth using realtime frame

interpolation for example if you

download a PDF on your phone one tap

we'll summarize it if you're browsing a

website on the internet you now have a

button that can read the entire page for

you and turn out a very good summary in

like 2 seconds if you have an AI feature

that can be pulled off completely on

your device it's likely to be faster

because it doesn't need to connect to

the internet more reliable cuz you can

use it literally anywhere on the planet

regardless of whether you're connected

or not and also safer since none of your

data is actually having to leave your

phone at all and so in time Google

assistant or Siri could simply be

updated with similar capabilities to the

R1 with neural Hardware chips on device

they'll be more powerful faster and

convenient another idea is that these

Standalone AI devices could just be

integrated into a watch that way you

don't have to carry around a separate

device so the unfortunate truth could be

that this new segment might be over as

soon as it started unless there's a

killer application I could be completely

wrong though Keen Tech enthusiasts and

tinkerers might find ways to make the i1

useful beyond anything I can imagine

right

now so how does it work well the device

works by using what rabbit are calling a

l or a large action model instead of the

traditional large language model the

main difference is that it can be used

for actually doing tasks instead of just

understanding linguistic context and

replying to a question if you remember

those AI agents built on top of chat GPT

last year it's kind of like those for

those interested the R1 L model is based

off something called neuros symbolic

programming this method combines newer

neural networks with older symbolic AI

which represents objects as symbols and

processes the information that way

neuros symbolic programming has worked

in narrow Fields like robotic Automation

and CAD software I'll bring into brills

to explain a bit more modern neural

networks train a model on lots of data

and predict answers using best guesses

and probabilities but symbolic AI or

good old fashion AI as it's sometimes

called is hugely different symbolic AI

requires no training no Mass amounts of

data and no guesswork it represents

problems using symbols and then uses

logic to search for Solutions all right

so I did a lot of research into neuros

symbolic AI uh talks and research papers

and I learned that it has the same

trappings and limitations of just

regular symbolic AI you essentially will

Define a bunch of if El statements and

instead of using humans to determine the

variables that go into these IFL

statements you'll use an AI to extract

features from a certain system system or

process and then do IFL statements on

that output but one of the biggest

limitations from neuros symbolic AI is

that you need domain experts to craft

the IFL statements with the good type of

biases but bias is unavoidable is this

the right task for the job I'm a little

skeptical but maybe they know something

that we don't know everything that

they're promising could be demonstrated

by releasing an app which is kind of

weird that they are going for Hardware

that costs only $200 with no

subscription fee when we all know AI API

calls at scale is not cheap but I mean

who knows maybe they know something that

we don't know and have funding that

we're not aware of this could

potentially just be a little nce Trend

like polaro cameras or maybe industry

shift like the iPhone I have my

skepticisms but only time will

[Music]

tell want to get a ride to the office

there's the app for that want to buy

groceries there's another app for that

each time you want to do something you

fumbles through multiple pages and

folders to find the app you want to use

and there are always andless buttons

that you need to click if we can make an

AI trigger actions on any kind of

interface just like a human word it will

solve the problem so how does the rabbit

manage to execute the apps that you have

without you doing

anything well there's a setup process

which means you have to log into the

apps that you normally use so it has

access to the data within you can also

train the R1 on new apps by using the

company's web portal and showing the

device how you use that particular app

and once it's being shown it can do that

automatically on your behalf this is

cool and opens up a lot of doors but it

seems like a lot of effort to the Casual

user who just wants the device to work

so what are some advantages it's fast

something that stood out to me is the

response time of a claimed 500

milliseconds or less that takes a lot of

the jankiness out of the experience so

that's very interesting unlike today's

voice assistance on a phone you don't

have to speak specifically today's voice

assistants will often get confused if

you ask them to do a task the R1 being

like chat GPT but also being more able

to execute tasks is much more useful but

probably one of the most interesting

aspects is the price costing just $199

with no subscription compared to the

Humane which is $600 with a subscription

it's compelling in this new

segment so what about the

disadvantages you have to carry around

another device that might not possess

the utility to Warrant doing that when

I've talked to friends and people about

this device everyone just says why can't

it just be an app being vo first can be

awkward and goes against current social

conventions there's also the issue of

complete confidence in the technology

take Uber for example will you trust the

R1 to do the task correctly if you're

booking your flights and hotels there

might be specific preferences which

could be missed because the device

simply couldn't know that also for

ordering food without seeing menu

options it seems like you miss out on

browsing on what you

want so when conclusion the rabbit R1 is

the archetype of a revolutionary product

but its usefulness is undetermined and

killer applications aren't yet apparent

in my view it's a great tool for early

adopters who love to Tinker but the

mainstream will take some time

especially getting used to a voice first

device but that's not the end of the

story many people do roll their eyes at

the term AI but you can't throw the baby

out with the bath water with any new

technology there's going to be scams and

gimmicks but also some useful

applications the question is what

category does the rabbit fall into it's

hard to say I think the rabbit is

interesting because it's bootstrapped to

the ever improving capabilities of AI

what do I mean by that well in theory

it's possible those 50,000 pre-order

customers who begin to show the rabbit

how to use their favorite apps

essentially become a huge aggregate

training data set for the lamb system

this could mean that the R1 eventually

becomes as good as humans at

understanding any app interface all it

would need is a software update when the

aggregate training is complete and that

that's fascinating to think about and

because of that along with a modding

Community I'm not so quick to throw this

concept away those were just my thoughts

I'll leave you with a question what

custom uses can you think of for such a

device getting a conversation going

around this could be really cool in the

end whether capable AI assistants come

from smartphones or a standalone device

Steve wnc was right after all eventually

all all those apps and all yeah I just

want to get to worry about not even

having to find an app you still have to

find the app app but I want the apps to

be able to become a part of the voice

system I want every program to start

coming out and being a part of Siri

every app on the iPhone and the same

thing would apply to other Tech other

platforms I just want to give a quick

shout out to nebula if you're tired of

YouTube ads and want to see all your

favorite educational creators like real

engineering half as interesting and W

over Productions in one spot nebula is

the place for you it's a streaming

platform built by creators for creators

you can can see my videos ad free and

there's exclusive content by some of the

best in the business there's also no

algorithms for us to worry about so we

can experiment for example polymatter

has done an exclusive video solely on

China's one child policy and it was very

interesting it's part of his larger

series called China actually if you sign

up using the link below you get to

support me directly and get nebula for

40% off an annual plan that's a little

over $250 a month it's the best deal in

streaming I'm looking to expand cold

fusion this year and nebula's expertise

and deep understanding of the challenges

of content creation will help with that

signing up to the link also helps me

achieve that goal so thank you if you're

interested in anything science

technology or business feel free to

subscribe to Cold Fusion it's free so

thanks for watching my name is toogo and

you've been watching cold fusion and

I'll catch you again soon for the next

episode cheers guys have a good

[Music]

one

[Music]

cold fusion it's newth thinking