0:00

Hello, and welcome to a

0:02

free preview of Sharp Tech. You've

0:08

made a brand promise and

0:11

you've built in capabilities to do

0:13

these sorts of things that you

0:15

have to rebuild from scratch. If

0:17

Brian uses his Alexa to turn

0:19

his lights on and off, guess

0:21

what the LLM -enabled version of Alexa

0:23

better be able to do? Right.

0:26

And better be able to

0:28

turn his lights on and

0:30

off, right? That's the thing, probabilistic

0:32

household chore maintenance doesn't sound

0:34

all that attractive, like deterministic solutions.

0:36

Right. Are they going to

0:38

turn on or off, right? It's

0:40

like shorting her's lights, right?

0:42

It's like, you're not going to

0:44

know until you try. And

0:46

so that's number one. They have

0:48

all these capabilities that were

0:50

built on the old way of

0:52

doing things that they have

0:54

to reimplement and they have to reimplement

0:56

to the same level of reliability.

0:58

They had before, when that previous level

1:00

of reliability entailed offloading a huge

1:03

amount of the complexity to the user.

1:05

And not just in a technical

1:07

perspective, but from a mental mindset perspective,

1:09

where I blame myself when it

1:11

doesn't work. And again, given what these

1:13

things were capable of, that was

1:15

a good place to have the product

1:18

where users had to do the

1:20

work. But now if they want to

1:22

do this, they have to redo

1:24

that. And that is even before then you

1:26

get to part two, which is just doing

1:28

this stuff at scale. Like how does the cost

1:30

structure work? Like right now you have a

1:32

free version of chat, GPT, but if you actually

1:34

want the decent functionality, have to pay the

1:36

subscription. Is there going to be a Lexa subscription?

1:38

Are you like, how are you going to

1:41

pay for the inference? How are you going to

1:43

do this at scale at cost? All the

1:45

obvious bits about doing an LLM, but I

1:47

think it's that first part that's

1:49

the most important. Is Alexa Plus going

1:51

to turn Brian's lights off consistently?

1:53

And it has to be able to

1:55

do that, which is a significant increase

1:58

in what... our

2:00

expectations of alleabs today. Part of

2:02

the whole thing with people's allebs,

2:04

they know they're gonna who listenate.

2:06

They know they're going to screw

2:08

up. That's very different from our

2:11

expectations of these these digital canisters

2:13

with disembodied voices inside of them.

2:15

Yes, no, exactly. I mean, it's

2:17

a real technical challenge. And the

2:19

verge adds that nobody in the

2:22

media got to try all the

2:24

new capabilities associated with Alexa Plus,

2:26

and they won't be able to

2:28

try the new capabilities until late

2:31

March at the earliest. A footnote

2:33

on the Alexa Plus page indicates

2:35

it won't be coming all at

2:37

once either. Some features represented are...

2:39

in development and will be released

2:42

in future updates was part of

2:44

the demo, which I think speaks

2:46

to the complexity that you're describing

2:48

there. And how tough it is

2:50

to crack. Yeah, no, exactly. An

2:53

Apple intelligence, I mean, we're months

2:55

into the Apple intelligence era and

2:57

we're still waiting for some of

2:59

the features that we saw in

3:01

the Apple intelligence commercials six months

3:04

ago. So all of this stuff

3:06

is difficult. Andy Jassie did say

3:08

that new beautiful hardware would be

3:10

coming this fall. A show is

3:12

very well known for it. Beautiful,

3:15

beautiful hardware. Exactly. Just channeling the

3:17

spirit of jobs. So Bob in

3:19

Michigan asks, is Alexa Plus too

3:21

late? Is this like the Windows

3:23

phone where network effects have taken

3:26

off an Amazon success with dumb

3:28

Alexa makes it hard for them

3:30

to make the jump? What do

3:32

you think about that? It's pretty

3:34

interesting because as a side note,

3:37

let's talk about Siri. I think

3:39

in the fullness of time Apple

3:41

probably wishes they had never watched

3:43

Siri and then today they could

3:46

launch Siri undergirded by an LM

3:48

and it's this incredible experience and

3:50

they could slowly build up the

3:52

Siri like all the capabilities that

3:54

are in Syria now, which doesn't

3:57

work as well as Alexa and

3:59

Google Voice, but the things like

4:01

turning on your lights and turning

4:03

those sorts of things. If you

4:05

could start from scratch and from

4:08

an Apple perspective, that would be

4:10

totally fine because the iPhone has

4:12

not been selling because of Syria

4:14

capability. It's been selling because it's

4:16

an iPhone, and so they could

4:19

start. Rolling out a voice assistant

4:21

in 2025 and be fine and

4:23

start from scratch with a completely

4:25

new sort of capability and Honestly,

4:27

maybe this is what they should

4:29

have done with Siri is have

4:31

an option That's like if you

4:33

want Siri as is That's one. We

4:35

have a new voice called Joey. I

4:37

don't know what everyone and you can

4:40

choose that options Joey will not be

4:42

able to do anything Siri does It

4:44

can have a good conversation,

4:46

it can't turn your lights

4:49

on and off. We are

4:51

going to add that over

4:53

time instead of this sort

4:55

of attempting to retrofit

4:58

Siri. Maybe that would have

5:00

been a better approach, but

5:02

they could do that because

5:05

they, the iPhones have sold

5:07

itself for years and people

5:09

buy iPhones for other reasons.

5:12

Right. Amazon does have a

5:14

lot of advice in their house.

5:16

From what I've seen anecdotally and

5:19

have to some extent sort of

5:21

read, the number one use case

5:23

is listening to music. And people really

5:25

like this. They like just asking Siri

5:28

to play stuff. It hurts my soul

5:30

to hear the quality of the music

5:32

that comes out of some of these

5:34

devices. Not Siri. great, but a lot

5:37

of people just don't care. And they

5:39

have this in their houses and maybe

5:41

they use it again. If they're a

5:43

little more advanced, they have it to

5:46

turn their lights on and off or

5:48

to open and close their curtains. And

5:50

that's sort of about it. And I

5:52

wonder if Amazon would be well served by

5:55

all the, all these integrations and all these

5:57

like bespoke use cases, they should

5:59

just like. dump all of them and

6:01

sort of start out. But they do

6:03

have an advantage that they have a

6:05

lot of devices in-house. They've sold 300

6:07

million to your point, 80 million are

6:10

used, or 80 million people are using

6:12

them, or however those numbers work out.

6:14

So I don't, there is a bit

6:16

where that is an advantage. They have,

6:18

if they can deliver something here, they

6:20

have a base to build off of.

6:23

So I can understand why it's taken

6:25

so long for Alexa plus to come

6:27

out. I think the problems are significant.

6:29

And I do think they have an

6:31

advantage because they do have these, but

6:33

they don't have to completely retrain their

6:35

users. So it's going to be, it's

6:38

going to be a really hard thing

6:40

to pull off. I don't think it's

6:42

necessarily a Windows phone sort of device.

6:44

Are people going to use Alexa Plus

6:46

instead of ChatGPT on their computer? Maybe,

6:48

maybe, I guess there's a possibility. It's

6:51

worth a shot. Yeah, when you lay

6:53

out all the challenges, I don't know

6:55

why they didn't rebrand it more aggressively.

6:57

There's Alexa and apparently Amazon users are

6:59

going to have the option to use

7:01

regular dumb Alexa, and if they don't

7:04

want to try Alexa Plus, and Alexa

7:06

Plus will be available without charge to

7:08

prime subscribers, $20 a month to everybody

7:10

else. But I just feel like there

7:12

are two. pretty different products. And so

7:14

I don't know why they didn't differentiate

7:16

the name more. But I do wonder,

7:19

I mean, it's still so early when

7:21

you look at this market and there

7:23

isn't really AI hardware in the market

7:25

that looks like it's gonna predominate for

7:27

the next couple of decades. So I

7:29

don't blame them for taking a shot

7:32

and saying, well. Maybe we're going to

7:34

be the ones that people are talking

7:36

to in the home since we already

7:38

have all these devices in the home.

7:40

I think it's worth pursuing, but I

7:42

have no idea whether Amazon is actually

7:44

going to be the long term UX

7:47

for AI for a lot of people.

7:49

But it's reasonable to say that it's

7:51

not going to be an app or

7:53

a website. I don't think that's crazy.

7:55

No, totally. I think You're totally right.

7:57

And I've long maintained, it's funny because

8:00

the original echo came out like four

8:02

or five months after the Amazon phone

8:04

came out and I dumped all over

8:06

the phone immediately. And that was even

8:08

before. reviewers got a hold of it

8:10

and realized he was one of the

8:13

biggest pieces of garbage they'd ever used.

8:15

And then so four months later they

8:17

come out the echo and everyone is

8:19

like prime for Amazon sucks at hardware.

8:21

They're stupid while they've been doing this

8:23

and they just sort of they took

8:25

all their phone criticisms and then applied

8:28

it and they just sort of they

8:30

took all their phone criticisms and then

8:32

applied it to the, because I think

8:34

they felt a little burn because we

8:36

were excited at first and they're going

8:38

to do another one. I love the

8:41

echo idea and the reason was this

8:43

is a real estate to try to

8:45

break into phones at 2014 or 2015

8:47

whenever this was was foolish. This is

8:49

the Windows phone. Windows phone already tried

8:51

and they'd failed and they had much

8:53

better a much better product than than

8:56

the fire phone. The home though, that's

8:58

the one place you don't necessarily have

9:00

your phone with you. Your phone has

9:02

to charge. You put it, you know,

9:04

hopefully you have some, a greater degree

9:06

of health. You know digital health than

9:09

I do and you you actually set

9:11

your phone down occasionally boundaries That's right

9:13

what we always I say that is

9:15

I do So the and so number

9:17

one it was it was number two

9:19

was a place that Amazon was associated

9:22

with you you buy stuff for your

9:24

house? So that angle sort of made

9:26

sense and number three Voice in talking

9:28

a voice interface in public has some

9:30

obvious problems with it who wants to

9:32

be out and about talking to their

9:34

device looking like a moron. And who

9:37

wants to hear people talking to their

9:39

voices and thinking, why am I stuck

9:41

next to this moron? In your house,

9:43

it's a private space. You can talk,

9:45

you can have a conversation, you can

9:47

ask it to play music. It just

9:50

seemed like if you're going to build

9:52

this. This was a bit of a

9:54

green filled opportunity and they grabbed that

9:56

and was the product good enough? Maybe

9:58

not. Maybe it was too early. But

10:00

it's just as good of a place

10:02

to take another swing at it. Again,

10:05

especially for a voice interface, the home

10:07

makes sense. They already have devices. The

10:09

device itself doesn't need to be that

10:11

capable. All it's doing is just passing

10:13

on. to a cloud service to do

10:15

the response, which Amazon is obviously well

10:18

suited to provide. So no, I like

10:20

them doing this. I think it's great.

10:22

And I'm not surprised it took so

10:24

long. And the execution is going to

10:26

be hard, but might as well. Why

10:28

not? Yeah, I don't know whether I

10:31

will be getting any of this new

10:33

hardware in my house, but if we

10:35

did I could imagine two use cases

10:37

one as my son grows up if

10:39

there's an AI that he can talk

10:41

to and ask questions to and get

10:43

you out of your parent duties Yeah,

10:46

I could see him using it. We

10:48

have a friend who's got like a

10:50

six or seven year old and he

10:52

loves LLLMs. It's just like the answer

10:54

machine. So if there's something you could

10:56

talk to absolutely and then also I'm

10:59

not an Amazon fan, but my wife

11:01

orders something from Amazon like a household

11:03

necessity. literally every day. I didn't know

11:05

every day or every hour. I wasn't

11:07

sure what unit of time you're going

11:09

to go in. Postpart of it's about

11:11

every hour these days. But so maybe,

11:14

I mean, those are two use cases

11:16

that would be accretive to Amazon's business.

11:18

Certainly the second use case. So time

11:20

will tell. Good luck. It's the technical

11:22

challenge. Good luck to all those guys.

11:24

Maybe one day we'll see a YouTube

11:27

video of this vaunted demo. But to

11:29

keep it moving, Christian, and the subject

11:31

line here was open AI is Blackberry.

11:33

A great. way to frame

11:35

good good take, all

11:37

he writes, that is the end

11:39

of the free right, and

11:42

that is the

11:44

end of the free

11:46

preview. If you'd

11:48

like to hear more

11:50

from Ben and

11:52

I, there are links

11:55

to subscribe in

11:57

the show notes, or

11:59

you can also

12:01

go to go to .fm. m

12:03

Either option will get

12:05

you access to

12:08

a personalized feed that

12:10

has all the

12:12

shows we do every

12:14

week, plus more more

12:16

great content from Sir

12:18

and the Stratechery Sir Techery

12:20

Check it out

12:23

and if you've got

12:25

feedback, if please email

12:27

us at please email us

12:29

at email at sharptech.org.