HomeVideos

The greatest AI tool ever??

Now Playing

The greatest AI tool ever??

Transcript

575 segments

0:00

OpenAI's quietly released what I believe

0:02

to be the biggest AI tool a major lab

0:05

has ever put out. I think it's not only

0:07

the future of AI, but the future of how

0:10

we interact with computers as a whole.

0:13

Now, this isn't hype. They're not paying

0:15

me. None of these major labs pay me or

0:17

even really talk to me at all. This is

0:18

based on my experience of using ChatGPT

0:21

voice over the last few days. It has

0:23

completely changed how I use AI. In this

0:27

video, I will show you how to do the

0:28

exact same thing. I will show you why

0:31

this has been an AGI moment for me, how

0:34

it's completely changed the way I use my

0:35

computer, and also give you a few

0:38

workflows you can use immediately to

0:40

100x your productivity. I'm not even

0:43

kidding. If you stick with me to the end

0:45

of the video, your entire life is going

0:47

to change. Now, let's lock in and get

0:49

into it. Let's first talk about how big

0:51

this moment is with ChatGPT voice. Let's

0:54

first talk about these AGI moments.

0:56

Moments where I used the technology that

0:59

completely changed how I view the world.

1:02

The first was Cursor with Sonnet 3.5.

1:04

This is when I really discovered vibe

1:06

coding. When I saw how much AI was going

1:09

to disrupt programming, this got me into

1:12

AI. This inspired me to start my YouTube

1:14

channel, all my content, everything like

1:15

that. Next was Claude code with Opus

1:17

4.5. This was the moment where I was

1:20

convinced software is basically solved.

1:22

AI can handle everything

1:24

software-related. Then there was Open

1:26

Claw, which really opened my eyes to AI

1:29

agent. It made me believe we would all

1:31

have our own army of thousands of agents

1:34

working for us in the near future. And

1:36

now is ChatGPT voice. This has opened my

1:40

eyes to ambient AI. What is ambient AI?

1:44

Ambient AI is the idea that there's

1:46

going to be an AI waiting for you

1:48

always, always listening, and can build

1:51

things for you no matter where you are

1:53

in the entire world. It completely

1:55

removes the barriers and the friction

1:57

between you coming up with an idea and

2:00

it being created and appearing in front

2:02

of you. Over the last several days of me

2:04

using Chat GPT voice, I have used it

2:07

everywhere. Walking around outside,

2:10

going on a hike, at a cafe, around my

2:12

apartment, in my bed, no matter where I

2:15

am, I can just say ideas that come to me

2:18

and they get built out and created on my

2:20

computer. Let me show you how this

2:22

works. So, up until now, the voice mode

2:25

for every AI tool ever made has been

2:28

dictation. What is dictation? You click

2:30

a button, you say things out loud, and

2:33

then they appear in your text chat, and

2:35

you hit send on it. See, there's a thing

2:37

I just said, you say things out loud and

2:39

they appear, then you hit send, right?

2:40

It's the same thing as using the AI

2:42

you've always used, you're just saying

2:44

it and then the text appears on your

2:45

screen. That's it, that's all dictation

2:47

is. This isn't dictation. This is

2:50

something completely different. This is

2:53

being able to command an entire army of

2:56

AI agents with just your voice. Let me

2:59

show you. So, if you hit this new button

3:01

in Chat GPT right here, start new voice

3:03

chat, it opens up Chat GPT voice.

3:07

So, I'll mute this real quick. We're now

3:09

in Chat GPT voice. You're probably

3:11

pretty familiar with this interface

3:13

before because they had a sort of

3:15

voice-like interface before where you

3:17

can chat with an agent, but now this is

3:19

superpowered. What makes this

3:21

superpowered? Now, your Chat GPT voice

3:24

agent can control any of your devices.

3:27

Any device you have that has the Chat

3:29

GPT app on it, your iPad, your iPhone,

3:32

your computer, your voice agent can now

3:34

go on any of them and control them. On

3:37

top of that, that agent can go and spin

3:40

up new agents and new threads on each of

3:42

those. So, basically, the power of your

3:45

voice can control all your technology,

3:47

get things built in an instant, and

3:50

explode your productivity anywhere you

3:52

are. And I'm going to demo this for you

3:54

in just 1 second and show you why this

3:56

is so different than anything else

3:57

you've ever used. Then go into the

3:59

workflow. But when you were using voice

4:01

before with AI, it was basically

4:03

dictation where you'd say things, it get

4:05

converted to text, the AI would read it,

4:07

then it would come up with text and

4:08

convert it to voice for you, right? It's

4:10

just a simple AI chat. Now it's an

4:13

entire command center. So watch how I

4:15

typically start my days. Hey chat GPT,

4:17

can you give me a status update on

4:20

School OS, on Personal OS, and on Henry?

4:23

Then give me a recommended next step

4:25

with each.

4:26

>> I'll check each project's current status

4:28

and then boil it down to one practical

4:30

next step for each.

4:31

>> So before when using voice AIs, it would

4:34

have access to just the chat. It can't

4:36

see other chats you're doing, other AI

4:39

agents you're using, other things going

4:41

on on your entire computer and all your

4:43

devices. Now because voice has access to

4:47

all your devices, all the chats,

4:49

everything all your different apps are

4:50

doing, it can go

4:52

>> School OS is deployed, but the course

4:55

sync is paused after an upstream error,

4:57

so no member course data is flowing.

5:00

Next is restoring a valid course

5:02

response and running on one member

5:03

canary before turning sync back on.

5:05

Personal OS is production stable,

5:08

including the video ideas release, but

5:10

Google calendar is waiting on your

5:12

approval at Google's security warning.

5:14

That's the next move, but the founder

5:16

attended three cycle production proof is

5:18

still undone, so no external beta yet.

5:21

The next step is to run that proof in

5:23

the controlled window and record the

5:25

verdict.

5:26

>> So I just got a status update on the

5:28

three biggest major projects I'm working

5:31

on right now. It checked out all the

5:33

chats, it looked at the GitHub to see

5:34

all the code updates, it looked at my

5:36

linear to see all the issues we have

5:38

queued at the moment. It can see across

5:40

everything. But here's where it can get

5:42

really interesting. Not only does it

5:44

have a view into everything you're

5:46

doing, all your chats, all your devices,

5:49

but it can spin up agents to work across

5:51

all of those. So, watch this right here.

5:54

Okay, can you spin up a new thread in

5:56

School OS to fix that error and get the

5:58

connector working? Inside the personal

6:01

OS, can you spin up a new thread that

6:03

makes the Google connect

6:05

that makes the Google Calendar connector

6:07

work and get that approved for me? And

6:09

then for Henry, can you actually build a

6:12

ChatGPT site that has a plan for how we

6:15

can do the entire production test and

6:17

run, and then send that all to me when

6:19

it's all ready and done?

6:21

>> Got it. I'm lining that up. All three

6:23

can be set in motion. I'll prepare and

6:25

verify the Google connection, and I'll

6:27

prompt you in a visible browser for the

6:29

one approval only you can complete. I'll

6:31

keep each effort isolated, so nothing

6:34

spills across repos.

6:35

>> And now, boom, look at this. Three new

6:38

chats spun up to do the work I just

6:40

approved. So, you can see the Henry test

6:42

plan's building a ChatGPT site for me

6:44

right here. There's a new chat, Personal

6:46

OS, it's fixing a calendar issue and

6:48

that's running for me. And School OS

6:50

started as well. So, it's spun up these

6:53

three new agents to begin work for me.

6:56

So, Voice is basically acting as my

6:58

chief of staff. I tell it what's on my

7:01

mind and it goes and it just delegates

7:03

the work to all the agents that need to

7:05

get it done. Now, you take it a step

7:07

further and you think about what this is

7:09

like when you start using it

7:10

cross-device. So, for instance, I was

7:13

out yesterday, I was going for a hike

7:14

for like 6 hours. I was on my iPhone,

7:18

and what you can actually do when you're

7:20

on your iPhone or your iPad or any other

7:22

devices, you can go into remote mode,

7:25

and when you're in remote mode, you can

7:27

actually start up your voice chat there

7:30

as well. And again, this voice chat has

7:33

control over all your devices. So, I'm

7:36

on the go, I have my AirPods in, and I'm

7:39

able to talk through all my work without

7:42

looking at a screen, without touching a

7:43

keyboard, without touching a mouse, and

7:45

get Agent spun up across all my devices

7:48

to keep all my projects moving forward.

7:51

I not only got the same amount of work

7:53

done that I typically do sitting at my

7:55

computer, I actually got more work done.

7:57

I'll tell you why it's so much more

7:58

efficient in a second, but I got more

8:01

work done walking around outside in the

8:03

redwoods of Northern California than I

8:06

do sitting at my desk with a keyboard

8:09

and mouse. And what's amazing is it's

8:11

ambient. And so you can give commands,

8:14

go on mute like I just did here, and

8:17

then walk around outside, sit at your

8:19

computer, read a book if you want, and

8:22

then when the work is done, ChatGPT

8:24

voice will just wake up again and say,

8:26

"Hey, by the way, we just finished this.

8:28

We're working on this now. This is

8:29

complete. You can check this out."

8:31

>> Personal OS calendar is built and

8:32

verified locally, including the daylight

8:35

saving fix. Google's approval page is

8:37

open, waiting on your consent. So

8:39

nothing's deployed yet. Do you want me

8:40

to push it live once you approve

8:42

Google's read-only calendar access?

8:44

>> Yes, please. You can do that once I

8:45

approve.

8:46

>> Got it. I'll handle the deployment right

8:48

after your approval.

8:49

>> And as you can see here, it just goes.

8:51

It's just ambient. I can be doing other

8:53

things.

8:54

>> once you approve Google's read-only

8:56

access, it'll go live and get checked.

8:58

>> And this is what makes it so great. It's

9:00

ambient. So while it's working, I can be

9:02

doing other things. I can be reading. I

9:04

can be working on other things. I can be

9:06

I can be going for a walk, and then it

9:08

just wakes up and says, "Hey, we just

9:10

completed this work. Here's the next

9:11

thing you need to do." And you can do

9:13

this on the go. I was sitting by the

9:15

pool yesterday as well. I had one

9:17

headphone in. I'm just sitting there

9:18

relaxing. Every once in a while my AI

9:20

wakes up, says something to me. I go,

9:22

"Okay, yes, continue. Okay, work on

9:24

this." And then it just goes. I'm

9:26

staying productive even when I'm walking

9:28

around, sitting by the pool. I mean, you

9:30

can look at me. I'm sunburnt all over

9:31

cuz I was spending so much time out

9:32

yesterday being productive. I'm going to

9:34

go into two workflows you can use

9:37

immediately in a second that I promise

9:39

will change how you work with AI

9:40

forever. Also, just feel free to jump

9:42

around down below if you want to the

9:43

different chapters. If you learn

9:45

anything at all so far, make sure to

9:46

leave a like down below, subscribe, and

9:48

turn on notifications cuz all I do is

9:50

make amazing videos about AI. Also, join

9:53

the Vibe Coding Academy. I'm doing a

9:55

live bootcamp on voice chat this week. I

9:59

will live show you way more workflows,

10:01

so make sure to join that down below. I

10:03

promise it'll be the best decision

10:04

you'll ever make. But before we go into

10:06

these workflows, let's talk about why

10:07

this has been such a massive

10:09

productivity gain and and all the

10:10

benefits you get from using this

10:12

workflow. Number one, you can't doom

10:15

scroll while conversating. If you're

10:16

anything like me, whenever you send a

10:18

prompt to an AI, it's so easy to kind of

10:20

just pop open Twitter or look at a

10:22

YouTube video and do things while the AI

10:25

works. And then all of a sudden it's

10:27

half an hour later and you haven't

10:28

responded to the prompt since you've

10:30

sent it. It is impossible to doom scroll

10:32

when you're deep in a conversation,

10:34

you're talking to an AI. I can't say

10:36

something and then it says something

10:38

back and I sit there and scroll for an

10:40

hour. You're forced to be engaged, so

10:42

you just get way more done and you're

10:43

way more focused. You can chat about

10:46

multiple projects at once. So this,

10:48

again, acts as your chief of staff. This

10:50

is how you want to think about ChatGPT

10:52

voice, which is it's your chief of

10:54

staff. It's not doing the work for you,

10:57

it's delegating your work to other

11:00

agents. So much like a CEO would have a

11:02

chief of staff that interacts with the

11:05

chief marketing officer, the chief

11:06

product officer, the CTO, it's the same

11:09

way where you're talking to it and it's

11:11

giving work to all your other agents. So

11:14

you're able to move the ball forward

11:17

with multiple projects at the same time.

11:19

When you're typing and you're texting

11:21

your agent, you can only text one agent

11:24

at a time and give one command at a time

11:26

for each project. So you're able to

11:27

multitask way better. You don't need to

11:30

be perfect, right? When you're typing a

11:32

text prompt out, you kind of need to be

11:33

perfect cuz it's going to read it very

11:35

literally. But when you're talking, you

11:38

can mess up, you can say um ahs, you can

11:39

talk about multiple things, and the AI

11:42

will truly sit there and think about

11:43

what it is you're saying, parse it, and

11:46

figure out the right things to do. And

11:47

this is maybe the biggest game changer

11:49

for me, and it can be done anywhere. All

11:52

of this can be done anywhere. My work

11:55

doesn't need to be here at this desk. In

11:57

fact, when I'm done filming this video,

11:59

I'm going to go outside and go for like

12:01

a 4-mi walk and just have one AirPod in

12:04

talking to my agent the entire time.

12:06

When ideas come to me, I'll say them out

12:08

loud. The agent will take it, assign the

12:10

task, and keep it moving.

12:12

>> Found the cause. Opening the second

12:14

connection over wrote the first task

12:16

secure callback.

12:17

>> me updates and everything we talked

12:18

about.

12:19

>> the newest one while signed in as

12:20

alex@100percentbetter.io.

12:25

>> I'm literally This is the first time

12:26

I've ever gotten work done while filming

12:29

a video. This is pretty amazing. So, I'm

12:31

going to show you three amazing

12:32

workflows with this right now, and then

12:34

I'm going to at the end right after this

12:36

give you a bunch of huge tips you need

12:38

to know so you can get the absolute most

12:40

out of this feature. Let's talk about

12:42

morning kickoff. I actually showed this

12:44

workflow to you at the beginning of the

12:46

video, which is this. You ask for a

12:48

status update on all your active

12:51

projects plus one step you can take to

12:55

move the project forward. Then once the

12:57

agent gives you the recommendation, you

12:59

go, "Okay, spin up a new thread for

13:02

every action you recommended." That's

13:04

the key here. That's one of the biggest

13:05

prompts you want to give when you're

13:07

talking is, "Okay, sounds good. Spin up

13:09

a thread to do that work." You never

13:11

want your voice agent to actually do any

13:14

of the work itself. You want it to

13:16

delegate. So, everything you do, say,

13:18

"Yep, okay, spin that off to a new

13:20

thread." And it will start a new chat in

13:21

whatever the appropriate project is, and

13:23

make sure that work gets done. Next

13:25

workflow is stream of conscious. This is

13:28

just when you go for a walk, put in your

13:30

airpod, and just start talking about

13:33

whatever's on your mind. Whatever you

13:35

think about the current projects you're

13:36

working on, your career, your goals,

13:39

your ambitions, what you're thinking

13:40

about, you just brain dump it all out to

13:43

your voice agent. It's just listening

13:45

the entire time as you're walking

13:47

around, and then you go, "Okay, based on

13:50

everything I just told you, what should

13:52

we build out? What can we build to get

13:54

me closer to my ambitions and goals?"

13:57

Then it will give you a bunch of ideas.

13:59

You go, "Okay, spin up a bunch of

14:00

threads to get all that done." And then

14:01

it goes, you come back to your computer

14:03

at the end of your walk, and all your

14:05

work is done for you, and you moved

14:06

everything closer to your goals. And

14:08

then the end of the day debrief, I like

14:10

to do this right before I go to bed. So

14:13

right before I put my phone on the

14:14

charger, I go, "I'm about to go to

14:17

sleep. What are three things we can do

14:19

tonight to push our project forward?"

14:21

Gives you a bunch of ideas. You say,

14:23

"Hey, spin up some threads to do that."

14:25

And it's good to go. Here's the key

14:26

thing. Here's the one connector between

14:29

all of these use cases and workflows.

14:32

You should be, when using ChatGBT voice,

14:35

asking way more questions than commands

14:39

you give. This is the beauty of having

14:41

super intelligence at your fingertips.

14:43

Like it's not really at your fingertips,

14:45

at your mouth. That's weird. It has way

14:48

more context, can think way faster,

14:50

knows way more things than you, no

14:52

offense. So you should be reverse

14:54

prompting as much as you can. "What

14:55

would you do here? How would you push

14:57

things forward?" The advantage here is

14:59

one, it comes up with ideas you never

15:01

would have thought of, but two, it

15:03

lightens your cognitive load. So if

15:06

you're walking around outside, you're

15:07

enjoying your day, you're talking to

15:09

ChatGBT voice, you don't want to, as

15:11

you're walking, have all these really

15:13

deep thoughts about what is every step

15:15

your agent should be doing next.

15:16

Instead, you go, "Hey, what do you think

15:18

we should do next?" It gives you a bunch

15:19

of ideas, then you just approve. That's

15:21

so much easier on your cognitive load

15:23

than having to manually think about

15:25

every single thing you want it to do.

15:27

So, you should be asking way more

15:30

questions. As long as you're using it

15:31

and saying, "Hey, what should we do

15:32

next?" You're going to get good results

15:35

from it. How do you improve that though?

15:37

How do you improve the recommendations

15:39

it gives you? Well, that's through what

15:40

I call a compass doc. So, one of the

15:43

first things you want to do in ChatGPT

15:45

voice when you boot it up is say, "Hey,

15:47

can you build a compass doc for each one

15:50

of our projects that explains to our

15:53

agents what the short and long-term

15:55

goals of this project are?" What ChatGPT

15:58

voice will do is create a markdown file

16:01

in each one of your major projects that

16:03

explains what you want to do with that

16:05

project in the long and short-term. Now,

16:08

when you ask, "Hey, what should we do

16:10

next on these projects?" your agent can

16:12

actually look at your compass doc and

16:15

figure out, "Okay, what are actions we

16:16

can take to reach that goal." Agents are

16:19

a lot more productive when they have a

16:21

goal to work for, not when it's just

16:23

some sort of arbitrary question, "What

16:25

should we do next?" If it has a goal to

16:26

work for, it can reverse engineer that

16:29

and figure out all the steps it should

16:31

take to get there. So, if you have it,

16:33

first thing you do, create a compass doc

16:35

for every project, you can now go

16:37

anytime you want to your voice agent and

16:39

say, "Hey, what should we do next?"

16:41

It'll make sure it creates a step that

16:42

gets you closer to the short and

16:44

long-term goals. So, here are my best

16:46

tips I'm going to give you on this. As I

16:48

said, number one, ask more questions

16:50

than prompts. You shouldn't be giving as

16:51

nearly as many commands as you give

16:54

questions. What should you work on next?

16:56

How would you handle this? What's the

16:57

best way to accomplish this goal? You'll

16:59

get way more better results and it'll be

17:01

easier on your brain to just ask

17:03

questions. Don't be afraid to rant or

17:06

mess up. I find that ChatGPT voice is

17:09

incredible at finding the unknown

17:11

unknowns, right? The things you don't

17:14

know that you don't know. So, if you go

17:17

and you just rant and you say everything

17:19

on your mind, Chad GPT voice can find

17:22

the angles that you haven't thought of

17:24

yet and find the solutions to those

17:26

angles. You also don't have to worry

17:28

about messing up. If you misspeak, it

17:29

can figure it out. You can say, "Mhm,

17:31

uh, uh." It'll be able to parse through

17:34

that and figure it out. They really

17:35

trained this voice model really well to

17:37

understand the true intentions of

17:39

everything you're saying. And then most

17:41

importantly, choose one device as your

17:44

headquarters. I have a bunch of devices.

17:46

I have three Mac Studios, an iPhone, an

17:48

iPad, a DGX Spark, a home PC. I have all

17:51

these devices. What you want to do is

17:53

this. You want to choose one of those

17:55

devices, preferably an always-on desktop

17:59

computer like a Mac Studio, a Mac mini,

18:01

or a PC and have that be your

18:04

headquarters. Then have every other

18:06

device be a node for those headquarters.

18:08

So, they control that main computer. So,

18:11

I'm on my Mac Studio right now. Because

18:12

this is my headquarters, when I go on my

18:15

iPhone and I give it commands, it'll

18:17

control and write the code on my Mac

18:20

Studio. Then if I go on my iPad and give

18:22

voice commands, it'll control everything

18:24

on my Mac Studio. So, it keeps all your

18:25

code and all your projects in one place.

18:28

This will allow you to go mobile and do

18:30

things a lot easier on the go. To set

18:33

all that up, you just go into your

18:34

settings. Then you go to connections

18:36

right here. And then on your main

18:38

headquarters computer, you click control

18:40

this Mac and then you add all the

18:42

devices that you want to allow to

18:44

control this computer. Once you do that,

18:46

you can go on your iPhone, your iPad,

18:48

wherever you are and you can have that

18:50

device from Chad GPT voice control your

18:53

main computer, which is amazing. And

18:55

again, it controls your entire computer.

18:58

You can say, "Hey, build a PowerPoint.

19:00

Hey, edit this doc. Open this up my

19:02

computer. Load this local model."

19:03

Whatever it is, you go into voice, it

19:06

controls it on your computer. This has

19:07

completely changed the way I work. The

19:10

The I'm done editing this video and post

19:12

it online, I'm going to be out in the

19:14

sun in the 80° weather of Northern

19:15

California walking around with my AirPod

19:18

in getting work done. I think this

19:20

completely changes the road map of every

19:23

AI lab. I mean, this is so powerful. I

19:25

think Anthropic tries to copy this as

19:27

quickly as possible. I bet SpaceX comes

19:30

out with a new kind of voice product

19:32

like this really, really soon. I

19:33

actually personally guarantee you that

19:35

the Jony Ive device that's going to come

19:38

out later this year is going to be some

19:40

sort of microphone product that ties

19:42

directly to this functionality. Maybe a

19:45

little pin or something that sits around

19:47

in your house that can hear you talk

19:48

anytime that you can just say prompts to

19:51

anywhere you are and it takes it and

19:53

does the work on your computer. I would

19:56

like bet anything that's what the

19:57

product is that's coming out. It's just

19:59

basically chat GPT voice but in hardware

20:01

form somewhere around your house. I have

20:04

over the last few days turned this

20:06

product on to like 10 of my friends and

20:08

all 10 of them were like, "Wow, this is

20:10

the most amazing thing ever. I thought

20:12

this was just voice dictation. This is

20:14

so much bigger than that." I promise if

20:16

you do the things I listed out in this

20:18

video, you take it, do it right now

20:20

after this video ends, I promise the way

20:22

you use AI will never, ever, ever be the

20:25

same. Hope this was helpful. You learn

20:26

anything at all, leave a like down

20:28

below. Let me know down in the replies,

20:29

have you used chat GPT voice yet? What

20:31

do you think? Subscribe, turn on noties.

20:33

I'll see you in the next video.

Interactive Summary

This video explores how the latest ChatGPT voice features represent a transformative 'AGI moment,' moving beyond simple dictation to become a powerful, ambient AI 'chief of staff.' The author explains how this tool can control multiple devices, manage complex workflows, and delegate tasks to specialized agents, significantly boosting productivity while enabling users to work effectively from anywhere.

Suggested questions

3 ready-made prompts