Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
Looking for a new ride? Shuff for your next car from
0:02
the comfort of your home, 100% online with Carvana. Carvana
0:06
has a massive inventory, including thousands of cars
0:08
under $20,000, so finding
0:10
a car that fits your budget and
0:12
lifestyle is hassle-free. Carvana makes car financing
0:14
hassle-free. Get pre-qualified in minutes by answering
0:16
a few quick questions. Carvana also offers
0:19
customizable terms and payments as low as
0:21
$0 down, because payment
0:23
plans should fit your plans. Visit carvana.com or
0:25
download the app today to find your next
0:27
car without ever leaving the comfort of home
0:30
or anywhere else. Terms and conditions may
0:32
apply. I had an interesting experience recently.
0:34
I went into a float tank for the first time
0:36
you've ever been in one of these? I
0:38
have seen them in sci-fi movies. Was
0:41
this like a back-to-tank from Star Wars where
0:43
you're healed from your injuries? No,
0:46
this is like a trendy new thing
0:48
in the Bay Area where you basically
0:50
go into these like pods. Imagine if
0:53
like Apple designed a coffin. It's like
0:55
a shiny white pod the size of
0:57
your body, and you
0:59
go in and it's filled with
1:01
a couple inches of very
1:03
salty water. So you
1:06
basically just lie there and you float
1:08
for an hour inside this pod, and
1:10
it's supposed to be relaxing. And was
1:13
it relaxing? Sort
1:16
of. I mean, it's definitely relaxing to be
1:18
like floating gently in some like warm salty
1:20
water, but it is a little
1:22
claustrophobic because you can make it totally dark in
1:24
there and do sort of sensory deprivation. But
1:27
the thing that actually made it less relaxing
1:29
for me was that you have music that
1:31
you can choose from, but the selection is
1:33
not good. It's like cheesy
1:35
yoga music. It's like pan flute. It's like,
1:38
you know, chimes. It's not what I wanted
1:40
to be listening to down there. Which
1:42
was espresso by Sabrina Carpenter. That's
1:45
what I want to be listening to. Kevin,
1:47
let me ask you this. Have you ever
1:49
heard of Taking a Bath? Because you're able
1:51
to get many of the benefits of driving
1:53
somewhere to float in two inches of water,
1:56
and you can choose your own music. Might want
1:58
to look into that. That's true. You know
2:00
you and I should go into into a pod
2:02
together and just start talking about tech news for
2:04
an hour You know what
2:07
they call that podcast I'm
2:13
Kevin ruse. I'm a tech columnist at the New York Times.
2:15
I'm Casey Newton from platformer and this is hard for Open
2:19
AI wanted Scarlett Johansson to be the voice
2:21
of chat CPT But then something got
2:23
lost in translation Then no one's
2:25
our bar the first person to get
2:27
Elon Musk's neural link implanted in his
2:29
brain Joins us to talk about how
2:31
a brain computer interface Changes Well
2:49
Kevin just as it seemed like
2:52
things were starting to settle after some wild
2:54
AI Demos last week
2:56
a shocking statement from one of the world's
2:59
most popular actresses has made us reconsider Everything
3:02
open AI has been telling us about its voice
3:04
assistant. Yeah, this is one of the craziest tech stories
3:06
of the year I've been totally obsessed with every twist
3:08
and turn. I'm very excited to talk
3:10
with you about it today Now did you ever think we
3:12
would have a literal Avenger fighting back against the relentless much
3:14
of AI? Is that sort of
3:16
what this story is about? So
3:18
last week we talked here about the
3:21
announcement from open AI about their new
3:23
GPT for Oh model Yeah, which was
3:25
most striking for this very flirty voice
3:28
assistant that they used in the demos
3:30
They showed us Kevin remind us what
3:32
was so striking about that demo So
3:34
the voice that they demoed it was
3:37
this sort of lilting female voice It
3:39
was a little flirty as you said
3:41
it sort of varied its register kind of
3:44
giggled at its own jokes it was like
3:46
it was it was very lifelike and realistic
3:48
and basically immediately as
3:50
the demo is going out people
3:52
start making comparisons to the movie
3:54
her and to Scarlett Johansson's character
3:56
in that movie Samantha and
3:58
like the the company itself made that
4:01
comparison, invited that comparison. Yes, Kevin.
4:03
And people are actually calling this
4:05
the greatest act of cultural appropriation
4:07
since Scarlett Johansson was cast in
4:09
Ghost in the Child. You
4:12
went there. That's right. So
4:14
on Sunday, OpenAI post to its
4:16
website this mysterious blog post titled
4:19
How the Voices for Chat GPT
4:21
Were Chosen. And in the
4:23
blog post, it says, quote, We believe
4:25
that AI voices should not deliberately mimic
4:27
a celebrity's distinctive voice. Sky's voice is
4:29
not an imitation of Scarlett Johansson, but belongs
4:32
to a different professional actress using her own
4:34
natural speaking voice. Kevin, when you saw that
4:36
blog post go up, did you have any
4:38
idea what was going on? No,
4:40
but it was one of those things where it's
4:43
like my my my sky is not an imitation
4:45
of Scarlett Johansson t shirt. It's
4:47
raising a lot of questions already answered by
4:49
my sky is not based on Scarlett Johansson
4:51
t shirt. It was like, if
4:54
you were saying this clearly something is happening in
4:56
the background, you did not just decide to come
4:58
out of this from nowhere. And it was
5:00
just a sign that things were going to get a
5:02
little weird. Yeah, absolutely. You know, this blog post went
5:04
up very late Pacific time on Sunday. And to me,
5:06
it was a sign that this was going to be
5:09
a rough night, which is the title of a 2017
5:11
film starring Scarlett Johansson. Okay, so
5:14
on Monday morning, things start to
5:16
become a little more clear when
5:18
OpenAI pulls Sky's voice from the
5:21
app. And Joanne Jang,
5:23
who is the model behavior lead at
5:25
OpenAI talks to the Verge and says,
5:27
quote, we've been in conversation with Scar
5:29
Joe's team. Pretty familiar there, Joey,
5:32
because there seems to be some confusion. We
5:34
want to take the feedback seriously and hear
5:36
out the concerns. And she
5:38
further suggested that maybe people hear similarities
5:40
because there are so few convincing female
5:43
voice assistants around. Does that seem convincing
5:45
to you, Kevin? No, of
5:47
course not. Because what? Well, you know, to
5:49
me, it just seemed like Scarlett was really
5:51
trying to get under the skin of OpenAI,
5:53
which, you know, under the skin is a
5:55
2013 film starring Scarlett Johansson. Oh, boy. All
5:57
right. So on Monday nights, Scarlett Johansson
5:59
herself released. There's a statement and this is
6:01
the do the right. Yes, it's sort of what
6:03
makes the world stops and to see really lays
6:05
out quite a marriage and was I think we
6:08
should walk through Yes so according the scarlet which
6:10
is why Color Sam. Altman had approached her
6:12
in. September Twenty twenty three about hiring
6:14
her to voice Sassy Be Teeth are saying
6:16
that it would be good for everyone to
6:19
see tech and creatives working together. A cabin
6:21
you remember that September was when they rolled
6:23
out voices to the In In a Taxi
6:25
Be T and In the up the rights
6:28
to around that same time it seems Sam
6:30
has this idea arm and you know to
6:32
me the seems. So
6:35
you're already employed by The Splits. Know
6:38
if you're going well girls are handsome Movie I
6:40
have every non medicines and or well I be
6:42
three It. Just seems clear that when same
6:44
approach or he wanted the prestige of
6:46
having her voice in the app the
6:48
press these being a two thousand six
6:50
I'm sorry for what amounts of aura
6:52
of so super entire I have really
6:54
attacks adequately a lot of a snow
6:56
at the time see declines for whatever
6:58
reason and then she writes in her
7:00
statements nine months later presumably referring to
7:02
last week's My friends, family and the
7:04
general public all noted how much the
7:06
newest system name Skies sounded like me
7:08
and how does she feel about it?
7:10
Well she said quote when I heard.
7:12
The release demo I was shocked, angered, and
7:15
in disbelief that Mr. Altman would pursue have
7:17
wasted sounded so eerily similar to my It's
7:19
the My closest friends and news outlets could
7:22
not tell the difference. Mr. Altman even insinuated
7:24
that the similarity was intentional Tweeting a single
7:26
word hurts and this really threw me for
7:28
a loops. I see says that two days
7:31
before the demo, Ahmed had reached out to
7:33
her egypt asking her to reconsider but then
7:35
open a I rolled out the demos this
7:37
guy before she could respond. so isn't that
7:40
a while? didn't know Crazy. So I mean
7:42
like. yeah we've got more about what it means
7:44
later but that this is the part where i'm just
7:46
like oh they they screwed this whole thing up so
7:48
badly yes and also like if there was a chance
7:50
that made you could work in a why wouldn't sue
7:52
you know wait for the and i think the answer
7:55
to that by the ways that they wanted to sort
7:57
of upstate google before it's a developer congress but to
8:00
finish out this statement, she says that she and
8:02
her lawyer sent two letters to Altman at OpenAI
8:04
asking for a detailed accounting of the process that
8:07
created the voice. And I think that is
8:09
probably what led to the blog post that went
8:11
up on Sunday night. And she closed with
8:13
a call to action. And I don't know if
8:15
we want to play the Star Spangled Banner underneath
8:17
this. I do think it would sound nice. I'll
8:20
just sort of read the quote in a time
8:22
when we are all grappling with deepfakes and
8:24
the protection of our own likeness, our own work,
8:26
our own identities. I believe these are questions
8:28
that deserve absolute clarity. I look forward to
8:30
resolution in the form of transparency
8:32
and the passage of appropriate legislation
8:35
to help ensure that individual rights
8:37
are protected. So after
8:39
that, Kevin, in the middle of the week,
8:42
OpenAI puts out a statement which they
8:44
attribute to Sam Altman. And it
8:46
says, quote, The voice of Sky is
8:48
not Scarlett Johansson, and it was never
8:50
intended to resemble hers. We cast the
8:52
voice actor behind Sky's voice before any
8:54
outreach to Miss Johansson. And to respect
8:57
for Miss Johansson, we have paused using
8:59
Sky's voice in our products.
9:02
So that was the statement in the middle
9:04
of the week. And I have spent the
9:06
last few days, Kevin, trying to figure out
9:08
what in the Vicky Christina Barcelona is going
9:10
on here. What did you find? Well, I
9:12
found first of all, that is a Scarlett
9:14
Johansson movie. I got that.
9:17
All right. All right. So on Thursday,
9:19
I had a chance to ask OpenAI
9:21
some questions. And my first question was,
9:23
who exactly at this company knew what
9:25
the heck was going on? Okay. And
9:27
what I was told was this, the
9:30
voice team decided they wanted to
9:32
record five voices for chat GBT.
9:35
But after that, they decided, hey, it
9:37
would be cool if we could get
9:39
Scarlett Johansson. And as part
9:41
of that, Sam Altman was sent out
9:43
on a mission to get Scarlett Johansson.
9:46
And according to them, that is when
9:48
he reached out to her in
9:51
September. And to sort of
9:53
bolster this timeline, they did a couple things. They
9:56
showed me a job posting from May of last
9:58
year where they advertised for Act for
10:00
these roles and I saw the job
10:02
posting. It did not mention Scarlett Johansson.
10:04
It did not mention her or any
10:07
other movies. They played... They did not
10:09
say only Black Widows may apply. No,
10:11
it didn't say that. Okay. They
10:16
then played for me a clip
10:18
from Sky's Audition where
10:21
she talks about, you know, walking around with
10:23
her toddler and basically
10:25
just gives you the impression of,
10:27
no, this is a real voice. This is
10:29
not a composite of other people's voices, which
10:31
is like one conspiracy theory that was sort
10:34
of floating around this week. And then finally,
10:36
they showed me a video clip of
10:39
the actor in the recording booth while
10:41
they were doing this recording. Now, this video clip
10:44
was very short. It was heavily pixelated and it
10:46
was taken from so far away that I couldn't
10:48
even tell where the human was supposed to be
10:50
until the second time I watched it. So
10:53
I wouldn't say that that clip alone is giving me
10:55
a lot of confidence in the narrative here, but I
10:57
have seen some sort of video suggesting that at some
10:59
point, a human being was saying
11:01
something into a microphone. Okay. So
11:04
let me just repeat all this back to you
11:06
and you tell me if I have the timeline
11:08
and the version of events, right? So OpenAI is
11:10
saying that they did not initially plan to have
11:12
a voice of Scarlett Johansson or
11:15
even one inspired by Scarlett Johansson as part
11:17
of this chat GPT voice release, but that
11:19
they later sort of came up with the
11:21
idea, well, maybe we should have
11:23
this sixth voice and maybe if Scarlett Johansson
11:25
will say yes, then we can get her
11:28
in as the sixth voice. That obviously never
11:30
happened, but they are basically saying this
11:32
was all never intended to
11:35
mimic the voice of Scarlett Johansson.
11:37
Any resemblance to people living
11:39
or dead named Scarlett Johansson is purely coincidental.
11:41
Is that basically what they are telling you?
11:43
That is what they're telling me. And
11:46
How do you feel about that narrative?
11:48
So Yeah, I guess I buy the
11:50
narrow version of events that OpenAI is
11:52
claiming happened here and I Also have
11:54
listened to clips of Sky and listened
11:56
to clips of Scarlett Johansson and they
11:58
don't sound totally accurate. Identical to me.
12:00
So it is totally plausible that they
12:03
had this other voice actor or play
12:05
this role. But there are so two
12:07
things that aren't quite adding up for
12:09
me. One of them is like, okay,
12:11
say you didn't. Mean. A Cast
12:13
a Scarlett Johanson sound alike. Why then
12:15
spend so much time around the lots
12:17
of this new voice teacher says making
12:19
people feel like they were listening to
12:21
Samantha from her to Serve Directly connect
12:23
the release of this product to this
12:25
movie and this actress. why do that?
12:27
If it you know it's going to
12:29
get you in trouble and then the
12:31
second thing is open. A I itself
12:34
has said in the past that they
12:36
do not want their synthetic voices to
12:38
serve mimic public figures. In fact, there
12:40
was actually a statement that they put
12:42
out on March. Twenty ninth. Earlier this
12:44
year in a blog post that Open
12:46
A I wrote called Navigating the Challenges
12:48
and Opportunities A Synthetic Voices And one
12:50
of the things they say in this
12:52
blog post is it There should be
12:54
a quote no Go Voiceless that detects
12:57
and prevents the creation of voices that
12:59
are too similar to prominent figures. I'll
13:01
be very interested if there is a
13:03
litigation around this issue. If you know
13:05
any of the discovery, they find evidence
13:07
that Open A I employees were served
13:09
talking about how similar his voice sounded
13:12
to Scarlett Johanson. Whether or not that
13:14
violated their own open a I policy
13:16
about not creating synthetic voices that were
13:18
too close to the voices a prominent
13:20
people. So let's talk about why this
13:22
matters because can understand you might be
13:24
listening. As saying the seems like kind
13:27
of a small saying, right? It's just
13:29
a voice Hates: If Super Intelligence is
13:31
coming soon, Is a voice really what
13:33
we should be worried about? Split? I
13:35
think it's important for a couple resist
13:37
and the first one to vince is
13:39
that the creative community is already deeply
13:42
skeptical of a. i rights last year
13:44
we have the sag aftra strikes and
13:46
this was a corpse length of the
13:48
fear of their which is that they
13:50
worried that companies would steal an actress
13:52
voice and image use it without their
13:54
permission as he and eventually either drive
13:56
them out of a job or just
13:59
it's right there wages way down. So
14:01
during that, thrite actors were able to win
14:03
concessions on this point. And now here you
14:05
have a different case where Scarlett Johansson wakes
14:07
up one day after saying no to this
14:09
company. And now the sound alike voice is
14:11
the voice of chat TBT. Yeah. So
14:14
Kevin, first of all, is what open AI
14:16
did here legal? It
14:19
may be it may not be. And it's
14:21
a little hard to tell. But I can
14:23
imagine that there's going to be some litigation
14:25
or maybe a settlement. I mean, Scarlett Johansson
14:27
has said that she has lawyered up. And
14:29
there are actually two legal cases
14:31
that people are sort of using to say
14:33
about this case involving Scarlett Johansson and open
14:35
AI that actually the facts would be highly
14:37
in her favor if she did decide to
14:39
litigate. No, do you want to tell us
14:41
about one? So one of them is a
14:43
case from 1988
14:46
called weights v Frito lay.
14:48
This is a case, the Frito
14:50
lay corporation makers of fine snacks decided
14:53
that they wanted to make a commercial for
14:55
a new flavor of Doritos. And
14:58
they really wanted Tom weights to sing
15:00
in it. But Tom weights is
15:02
sort of famously anti commercial, like he just
15:04
didn't want to have his songs used to
15:06
endorse products. It was like against his values.
15:09
So they went out and they paid $300 to
15:11
a Tom weights impersonator. Basically,
15:13
a guy who's in a band who
15:16
sounds exactly like Tom weights, and
15:18
they put the impersonator in the
15:20
Doritos commercial. They really
15:22
said weights, weights, don't tell me that you don't want
15:24
to be in this commercial. Yeah. Yeah.
15:27
So then the commercial comes out, Tom weights, his friends
15:29
start calling him and saying, Hey, I thought you were
15:31
against commercials. Why are you all of a sudden endorsing
15:33
Doritos? He gets mad, he
15:35
sues, and he sues not for copyright
15:38
violations, because you can't copyright a sort
15:40
of way of singing or talking, but
15:42
for false endorsement. And a
15:44
jury awards him $2.6 million. He's
15:47
basically the heir to the Cool
15:49
Ranch Dorito fortune. Congratulations. The
15:52
other one is around the same time, Bette Midler
15:54
had a very similar thing happen to her Involving
15:57
the Ford Motor Company, which went to her.
16:00
Her and said hey, could we use your
16:02
your song, your voice in a commercial for
16:04
our new Mercury Sable be the wind beneath
16:06
our wings But Midler? Exactly. So she says
16:08
no. I don't really want to endorse products,
16:11
so instead they go. and they hire one
16:13
of her former backup singers and basically instructors
16:15
backup singer to sound as close to bet
16:17
Midler as possible. See takes us to court
16:19
see when. Four hundred thousand dollars in damages?
16:22
kinds of I think Scarlett Johanson has and
16:24
not at and I would never say an
16:26
airtight case here to there's no such thing
16:28
spreading. She has a very. Strong case here.
16:30
split. but even setting aside the the legality
16:33
of it, Kevin, I'm curious to get your
16:35
thought on what this does for public perception
16:37
once. where are we right now and how
16:39
average people are thinking about ai, what role
16:41
at my play in their life and and
16:43
whether it might threaten them in some ways.
16:45
And you think this is the most so
16:47
damaging thing to come out of this particular
16:49
episode? It's not. Actually, you know I'm I'm
16:51
sure they will. You know I'd be will
16:53
figure out a way to sort of make
16:55
things right with crowds or Hansen where they'll
16:57
go to court. But I think the broader
16:59
damn as here. Is to the the public trust
17:01
in open A I This is a company that
17:03
has said he A we are building something that
17:05
was eventually become an artificial general intelligence. We're doing
17:08
this for the good of humanity and we want
17:10
you to trust us on that. and I think
17:12
they got the benefit of the doubt for a
17:14
while because they were releasing things that were cool
17:17
and useful to people. Taxi be teeth you know
17:19
with of a moment where a lot people said
17:21
okay maybe they are I'm in a at at
17:23
the from head of the pack here and maybe
17:25
we're okay with that then I think we've we
17:28
saw things start to decay a little bit with.
17:30
each successive release and the and the sort
17:32
of overall vibe being that this was actually
17:34
not sort of the a nonprofit research lab
17:37
as it had been started but was actually
17:39
something more like a very traditional tech companies
17:41
and so i just think we've seen a
17:43
gradual erosion of that process from the public
17:46
in open a eyes and i do ultimately
17:48
think that hurts them long term i'd i
17:50
agree with you i think this has been
17:52
a really bad a month for the perception
17:55
attack amongst average people i think this is
17:57
a moment where we have seen tech companies
17:59
get really greedy and greedy at
18:01
the expense of working people. And so
18:03
like as May is coming to a
18:06
close, on one side of the ledger,
18:08
you have Scarlett Johansson and an entire
18:10
creative class of workers rallying around her.
18:12
And on the other hand, you have
18:14
OpenAI Soundalike voice, Google AI overviews eating
18:16
the web and the Apple hydraulic press
18:19
from the commercial like crushing everyone into
18:21
a fine pulp, right? So I think
18:23
the tech industry needs a better story
18:25
to tell here than we're coming for
18:27
your voice and there's nothing you can
18:30
do about it. Totally. So there's a
18:32
second thing that I wanna talk about
18:34
though, Kevin, which is the implications of
18:36
this story for OpenAI because I think
18:38
it recontextualizes one of last year's biggest
18:41
stories, which was Sam Altman temporarily getting
18:43
bounced out of the company. So can
18:45
you just remind us what happened in
18:47
November to Sam? Yeah, so he
18:49
was fired in a surprise
18:52
move by a number of members
18:54
of the nonprofit board that governs
18:56
OpenAI along with Ilya Sutskover
18:59
who was the chief scientist at the time.
19:01
And by way of explaining why they
19:03
were firing him, they made these kind
19:05
of vague statements about how Sam had
19:07
not been quote, consistently candid. And
19:09
just basically implied that he was sort of
19:11
a slippery person who was
19:13
telling different things to different people and
19:16
who they had sort of lost faith
19:18
in. But they never sort
19:20
of gave many concrete examples
19:22
of that. And so I think it
19:24
was tough for people to understand why
19:27
make such a sudden and
19:29
important decision in sort
19:32
of the dead of night without consulting
19:34
anyone. And so Sam sort
19:36
of had the trust and the faith
19:38
of OpenAI's employees. And so they
19:40
rallied around him. They all, remember they were
19:42
gonna briefly go, I'll go work at Microsoft.
19:45
And then the board members end up
19:48
being sort of pushed off the board. And
19:51
Sam is brought back as the CEO.
19:53
It's a wild story, but I think
19:56
a very interesting thought experiment for me over
19:58
the past week has been. What if
20:00
the board coup had happened now? What if the
20:02
board had waited to make its move on Sam
20:04
until now, when I think, I think
20:07
it's fair to say he would not get the same benefit
20:09
of the doubt from the employees or the investors in OpenAI
20:11
that he has today. I think that's true. Now, this is
20:13
a company that is valued at what? 80 to $90 billion.
20:17
I think the employees who are working
20:19
there want to see the equity that
20:21
they have in that company realize and
20:23
I think that there's a very good
20:25
chance that if what you just
20:28
laid out happened, those employees
20:30
would still support Sam. That said, you're
20:32
right. We would have a good example
20:34
of why this board might be a
20:37
little bit concerned. I know at least
20:39
for me, ever since, you know,
20:41
as we did our own reporting, we talked to people
20:43
involved in that situation. I have
20:45
thought since November, there might be another
20:47
shoe to drop here, right? We may
20:50
eventually learn what the board was so
20:52
concerned about. And I feel like this
20:54
week for the first time, we actually
20:56
know now, like it is this sort
20:58
of thing. Yeah, but I actually, as
21:01
strange as it's gonna sound, I don't think
21:03
this Scarlett Johansson voice thing is actually the
21:05
worst thing that has happened to OpenAI over
21:07
the past couple of weeks. What do you
21:09
think it was? So there was this other
21:11
story that came out just recently about these
21:13
employee agreements at OpenAI. And this
21:15
came to light after Ilya
21:17
Sutskover and Jan Leikey, who were the heads
21:20
of the company's super alignment team, both
21:22
announced they were leaving the company. And
21:25
people started looking into the paperwork
21:27
that OpenAI employees have to sign
21:30
when they leave the company. And recently,
21:32
Kelsey Piper at Vox reported that there
21:34
was a really unusual provision in this
21:36
exit paperwork that basically said that if
21:39
OpenAI people left the company and then
21:41
spilled the beans or said something or
21:43
disclosed something about the company or disparaged
21:45
it in any way publicly, they
21:48
could not only break this contract,
21:50
but they could have their vested equity clawed
21:52
back, which we should explain why that's such
21:54
a big deal. So normally, you go work
21:56
at a tech company, you get stock options,
21:58
some of those stock options. options vest over
22:00
time. And this is traditionally how tech
22:02
employees make a lot of money. Their
22:05
stock options vest, they sell them, they
22:07
get the money. So
22:09
it is not unusual when you leave a
22:11
tech company to have your unvested equity forfeited.
22:14
What is extremely unusual, and actually I've never heard
22:16
of this happening before in the tech industry, is
22:18
for a company to say, we can actually take
22:20
back your vested equity if you've left the company
22:23
and you despair just publicly. So this was something
22:25
that a lot of OpenAI former
22:27
employees had been terrified of. It's a reason why we
22:29
haven't seen a lot of former OpenAI employees
22:31
speaking out. And when it
22:33
became public, a lot of people in the
22:35
AI industry said, this is crazy. We
22:38
have not seen this at any other companies.
22:40
They are trying to silence former employees for
22:42
speaking out. And then you saw actually Sam
22:44
Altman make this statement about it,
22:47
where he basically said, I didn't know about this. I
22:49
didn't know this was part of our paperwork. I've been
22:51
trying to get Scarlett Johansson to do the voice of
22:53
Chef GPT, the guy at time for this. So
22:56
he said, we've never clawed back anyone's vested equity.
23:00
Basically he said, this is one of the
23:02
few times I've been genuinely embarrassed running OpenAI.
23:04
I did not know this was happening and
23:06
I should have and we'll fix it. So
23:08
the reason I think this is actually a bigger
23:11
deal than the Scarlett Johansson thing, despite it not
23:13
getting nearly as much attention, is because OpenAI is
23:15
in a talent war, right? They are constantly
23:17
trying to pick off the best AI
23:19
researchers from all of the biggest companies
23:21
in Silicon Valley. It's a very talent
23:23
heavy business and talent dependent
23:25
business. And if they start losing
23:27
people who say, this company seems like they're slippery
23:29
and I can't trust them, I think that is
23:31
an existential threat to them in the long term.
23:34
Well, look, if
23:36
you're wondering why you've never heard me give an interview
23:38
where I talk about what it's like to work with
23:40
Kevin, I have signed something very similar. But
23:43
you're right, it is very unusual, particularly for
23:46
a company with Open in its name. And
23:48
I agree this was another black eye for
23:50
the company this week. It also of course
23:52
came on the heels of their entire super
23:54
alignment team being dissolved, which we discussed last
23:56
week. So there's just kind of a lot
23:59
of swirling drama. around that company. Now
24:01
we should also say while all that's going on,
24:03
OpenAI's business is doing great. I don't want to
24:05
pretend that it's not. Altman was on stage at
24:07
a Microsoft Developers Conference this week, which we'll talk
24:10
about in a little bit here.
24:12
There has been some reporting that Apple and
24:14
OpenAI are going to announce a big partnership
24:16
next month at Apple's own developer conference. And
24:18
finally, there was reporting this week that on
24:21
the day that the GPT40
24:25
model was announced, OpenAI's revenue shot at more
24:27
than 20%, according to
24:29
third party estimates. So clearly, OpenAI
24:31
is doing great. But
24:33
the sorts of things that
24:35
we've seen this week have given me some
24:38
pause. And I wonder if they've given you
24:40
some pause as you think of what is
24:42
the future of this company? Yes, I think
24:44
the correct way to phrase what I've been
24:46
feeling this week would be vibe shift. I
24:48
think there has been a big vibe shift
24:51
around AI when it comes to the creative
24:53
community, but especially with OpenAI as it relates
24:55
to the sort of trustworthiness of what they're
24:57
building. And I've talked to
24:59
people who say, I basically gave this company the
25:01
benefit of the doubt, I gave San the benefit
25:03
of the doubt, they seem to be
25:05
saying a lot of the right things. And now they're just
25:07
kind of like, I don't know, man. And I also think
25:10
it's like I was thinking about this
25:12
sort of idea of the
25:15
Silicon Valley builder's mindset of
25:17
like, ask for forgiveness,
25:20
not permission. And I think that's been
25:22
the way that a lot of successful
25:24
companies have been built in Silicon Valley.
25:26
Uber, Facebook, to some extent was
25:28
a story of asking for forgiveness, not permission.
25:31
And I think that that works with most technologies.
25:34
But I think with AI, it's a little bit
25:36
different. Yeah, absolutely.
25:39
It's also he asked for permission,
25:42
not for forgiveness, he did it backward.
25:44
He asked for permission, didn't get
25:46
permission, and then asked for forgiveness after he
25:48
didn't get permission, which is not something that
25:50
he didn't even ask for forgiveness. He said
25:53
I would never ask for forgiveness because there
25:55
was nothing to apologize for in the first
25:57
place because this voice isn't based on Scarlett
25:59
Johansson. It is the wild. sort of, you
26:01
know, hardest to pin down narrative. It
26:03
is. I mean, look, the thing that this brings to
26:05
mind for me, Kevin, is that you and I both
26:08
covered the decline in public perception around Facebook, right? And
26:11
Facebook once seemed like this silly little toy. Nobody
26:13
paid too much attention. Then after the 2016 election,
26:15
everybody is like, wait, is this secretly a mind
26:17
control device that is, you know, making all of
26:19
our teenagers insane and everything else? And
26:21
obviously, it's way too soon to say that something like
26:23
that is happening to OpenAI. But
26:26
I'm telling you, this is how it starts, right? I
26:29
think more and more people are becoming convinced every
26:31
day that whatever AI is going to be in
26:33
the short to medium term is not going to
26:35
be a good bargain for them. And they're not
26:37
going to give OpenAI the benefit of the doubt.
26:39
And that means that OpenAI, I think, needs to
26:41
be really careful in how it makes its next
26:44
several decisions around this kind of stuff. So,
26:46
look, I'm glad the Hollywood A-list is paying attention.
26:49
If you are worried about what a company like
26:51
OpenAI might do with your voice or your job,
26:53
you're in good company, which is a 2004 movie,
26:56
Starling Scarlet. Oh, good. I was
26:58
worried there was one Scarlett Johansson movie we weren't going
27:00
to mention. When
27:03
we come back, Neuralink's first ever patient
27:05
joins us to talk about how the
27:08
technology is changing his life. Looking
27:26
for a new ride? Shop for your next car from the
27:28
comfort of your home, 100% online with Carvana. Carvana
27:32
has a massive inventory, including thousands of cars
27:34
under $20,000. So, finding
27:36
a car that fits your budget and
27:39
lifestyle is hassle-free. Carvana makes car financing
27:41
hassle-free. Get pre-qualified in minutes by answering
27:43
a few quick questions. Carvana also offers
27:45
customizable terms and payments as low as
27:48
$0 down because payment plans should fit
27:50
your plans. Visit carvana.com or download
27:52
the app today to find your next car
27:54
without ever leaving the comfort of home or
27:56
anywhere else. Terms and conditions may
27:58
apply. Hi, I'm Megan
28:00
Loram, the director of photography at The New
28:03
York Times. A photograph can
28:05
do a lot of different things. It can
28:07
connect us. It can bring us to places
28:09
we've never been before. It can
28:11
capture a story in a universal visual
28:13
language. But one thing that all these
28:16
photographs have in common is that, you
28:18
know, they don't just come out of
28:20
the ether. We spend a lot of
28:22
time anticipating news stories, working with the
28:24
best photographers across the globe. These
28:26
are photographers who have spent years
28:28
mastering their technical craft, developing
28:30
their skills as visual chroniclers of
28:32
our world. You know, getting certified
28:34
as a scuba diver and learning
28:36
how to shoot underwater to document
28:38
climate change or tremendous cardiovascular training
28:40
in order to ski on the
28:42
slopes next to Olympic athletes. This
28:44
is an effort that takes tons
28:46
of time and consideration and resources.
28:48
All of this is possible only
28:50
because of New York Times subscribers.
28:52
If you're not a subscriber yet,
28:54
you can become one at nytimes.com
28:57
slash subscribe. So,
29:01
Casey, as you know, one of the
29:03
technologies that is fascinating to me right
29:05
now is the Brain Computer Interface or
29:07
BCI. That's right. In fact, it's so
29:09
fascinating to me that I forced you
29:11
to try one a few weeks
29:13
ago and I think it's safe to say it was not
29:15
all that impressive. No, I was willing to go through with
29:17
it because I've always wanted to see if we could detect
29:20
any brain activity for you. But afterwards
29:22
I thought, I can't do this anymore.
29:25
So, just to back up, Brain Computer
29:27
Interfaces are a type of technology that
29:29
allows you to basically control a computer
29:31
directly with your brain. And this has
29:33
been something that obviously Science Fiction has
29:35
been talking about and that scientists have
29:38
been working on for decades. The company
29:40
that is perhaps best known in this
29:42
space is Neuralink, Elon Musk's Brain Computer
29:44
Interface Company. And I've just been
29:46
fascinated by this whole area. I
29:49
made you try one. It was what's called a
29:51
non-invasive brain computer interface, which means it's not inside
29:53
your literal head. It's like a headband that you
29:55
wear. And it doesn't work that well. The technology
29:57
was not all that impressive. We didn't end up...
30:00
airing the segment where we tried this thing on because
30:02
it just wasn't very good. But
30:04
with Neuralink, the implant, the brain computer
30:06
interface goes literally inside your skull on
30:08
your brain and it allows you to
30:11
control a cursor with your mind. So
30:13
I think we should tell people a
30:15
little bit kind of about like what
30:18
this thing is and what it looks
30:20
like because there are these threads that
30:22
have electrodes on them that penetrate into
30:24
the brain and those electrodes
30:27
read signals which then get translated
30:29
through the Neuralink device and that's where
30:31
I've lost the plot. Can you pick it up from there?
30:34
Yeah, so it basically translates your electrical activity
30:36
in your brain into commands to control something
30:38
on the outside of your body like a
30:40
computer or something like that. And
30:43
for years now, tech companies have
30:45
been looking at using BCIs to
30:47
help people who have debilitating conditions
30:50
like a spinal cord injury or
30:52
a stroke or some limitation on
30:54
their mobility. But also a
30:56
lot of people in Silicon Valley just talk about
30:58
this as a potential next step in computing altogether.
31:01
Like in the future, some or
31:03
all of us will have these kind of brain computer
31:05
interfaces. And I've even
31:07
heard people in the AI world say, this is
31:10
the way we are going to stay ahead of
31:12
the robots as the AIs get smarter is that
31:14
we are going to implant computers in our head
31:16
that will basically increase our own cognitive capacity. That
31:19
technology does not exist. It is mostly just an
31:21
idea from science fiction. We have no idea whether
31:23
that would work or not. But this
31:25
is a big important technology that a lot of
31:27
people in tech are excited about. And
31:29
just over the last few months, we've actually seen one
31:32
of the clearest views yet into how this might
31:34
work in a real human. Yeah. And
31:36
while you've just spun out some really fantastical
31:38
sci fi scenarios, Kevin, what appeals to me
31:41
about this story is that it is a
31:43
case of technology helping one person who had
31:45
something really terrible happen to him. So I
31:47
want to say up front, very few people
31:50
have been as critical of Elon Musk as
31:52
I think the two of us have been
31:54
on this show in particular. And
31:58
as hard for me as it is to set aside. my
32:00
personal feelings particularly about what he did to Twitter.
32:03
I truly am so inspired by how
32:05
this technology is helping one person and
32:07
I think it is absolutely worth understanding
32:10
what is this thing that they built
32:12
and how has it changed the life
32:14
of at least one individual. Yeah so
32:17
today we have a really special opportunity
32:19
to talk to the one person on
32:21
earth who has actually gotten the Neuralink
32:23
brain computer interface implanted in his skull.
32:26
This man is named Nolan Arbaugh and
32:28
he now has a computer system about
32:30
the size of a coin and a bunch of
32:32
threads with electrodes that connect to his brain that
32:35
allow him to do things like move a
32:37
cursor around on a computer with
32:39
just his mind and this is a
32:41
big deal for Nolan because for the
32:44
last eight years he's been paralyzed from
32:46
the shoulders down he had a freak
32:48
accident eight years ago where he suffered
32:50
a severe spinal cord injury and so
32:52
he has been a quadriplegic for the
32:54
last eight years and he volunteered to
32:57
be patient number one in this Neuralink
32:59
experience. So today we're going to talk to
33:01
Nolan about what having this Neuralink device in his
33:03
brain has been like how it's changed his life
33:05
and why he volunteered to risk his body on
33:08
this unproven new technology. Let's bring him in.
33:15
Nolan Arbaugh welcome to Hard Fork. Nice to
33:17
be here thanks for having me. Yeah where
33:19
are we catching you right now? Just describe
33:22
where you are. I'm
33:24
in my house in my
33:26
bed so if you hear any weird noises
33:28
in the background it's my bed I have
33:31
an air mattress so it's kind of blowing air through the
33:33
whole thing all the time. Got it. Where's
33:35
home for you? Yeah I'm in Yuma, Arizona. Cool.
33:39
So it's been a crazy last
33:41
few weeks for you back in
33:43
January Elon Musk announced that the
33:45
first human patient had been successfully
33:48
implanted with a Neuralink device but
33:50
he didn't say the name. It's
33:52
only really in the last two months that your
33:54
name has become public, and just the last week
33:57
or so that you've started to talk more broadly
33:59
about your experience. What is it been
34:01
like to be served? the literal face of
34:03
this technology? Yes, But all
34:06
right I'm just here trying to
34:08
get all the information out as
34:10
many people as possible. I think
34:12
it's amazing technology. I think what's
34:14
going on in my life and
34:16
what I for see the future
34:18
will hold is worth bringing. The
34:20
whole world along with. Such
34:23
been caught man own. I'm enjoying it so
34:25
far. Yeah, I
34:27
just want to go back to
34:29
serve before you got this narrowly
34:32
device implanted in your head. What
34:34
compelled you to register to participate
34:36
in this extremely your new and
34:39
untested experiment? Yeah. I
34:41
mean I didn't really know much about it.
34:43
My buddy called me up one day and
34:45
kind of gaming the five min a rundown.
34:47
I wasn't expecting anything the come from it
34:49
so I'll admit a few joke. I'm application
34:51
you know I just I figured I would
34:53
never your back and then once I did
34:55
start hearing that I had to think about
34:57
a little bit more. Seriously had very serious
34:59
and candid conversations with my parents, my friends.
35:02
And. Ultimately when it
35:04
came down to it and are selected
35:06
I decided that. I just wanted to
35:08
help. I knew the I wanted to help make it
35:11
safer for everyone after me and I knew that I
35:13
wanted her. You. Know try to make
35:15
a difference in the world and something I've
35:17
been trying to do something of I'm looking
35:19
for for a years and the seem like
35:22
the perfect opportunity. Dt.
35:24
Remember any of the jokes the
35:26
put on your application? Yeah, I
35:28
said I wanted an Iron Man
35:30
suit. I said that I wouldn't
35:33
mind being uploaded in the Matrix.
35:37
You know, seems like a cigarette and
35:39
other a one of the truth is
35:41
that case it I have learned in
35:43
reporting about Tech over the past decade
35:45
is that you never want to try
35:47
version One point and right, it's risky
35:49
to try the first version of anything
35:51
because the bugs are still being worked
35:53
out usually. there's some some rough edges
35:55
of split you are literally being asked to
35:58
try version one point out not just a
36:00
new gadget but of something that is going
36:02
to go inside your skull. So was that
36:04
part of your process of thinking through this?
36:07
No one has ever had one of these put in their brains
36:09
before. Maybe I want to let someone else
36:11
be the first person. Yeah,
36:14
it crossed my mind. Something that my
36:16
buddy and I, the buddy that called
36:18
me on the phone, we talked about
36:20
at length was, you know, this is the worst
36:22
version of it that's ever going to be in a human. Maybe
36:26
someone else should go first. And
36:28
I'll get a better version later on down the road. Or maybe I
36:30
don't do it at all and wait for it to be on
36:32
the market to the public and then I get an even
36:34
better version. But ultimately I figured
36:37
that if anyone's going to do it, then
36:39
I should. I have a pretty
36:41
solid foundation with my faith in God. And
36:44
I just felt like I've thought my
36:46
entire, you know, accident. I'm glad that it
36:48
happened to me and no one else I
36:50
know because it's just a very hard thing
36:52
to experience, you know, being a quadriplegic. And
36:55
I wouldn't ever want any of my friends to have to
36:57
go through this. So it's just a mindset that I've had
36:59
forever. And with the Neuralink,
37:02
it was the same thing. Like if anything were
37:04
to go wrong, I would feel terrible if I
37:06
passed up to wait for a better version and
37:08
something went wrong to someone else. So I
37:10
knew that it had to be me. Your
37:13
parents are your primary caregivers. What were
37:15
your conversations with like them about doing
37:17
this? Yeah, they were really
37:19
hard for a while. I
37:21
mean, as a quadriplegic, the only real
37:24
thing that I have left is my mind,
37:26
is my brain, is my personality. And
37:29
it's hard to let someone go in
37:32
and kind of rummage around up there,
37:34
especially with something that's never been tested
37:36
in a human. So one
37:38
of the things that I mentioned
37:40
to them was that if I had
37:42
any sort of brain deficiency afterward, if I was
37:45
mentally handicapped in any way, that I didn't want
37:47
them to take care of
37:49
me anymore, that I wanted them to put me in
37:51
some sort of home. Because taking care of a quadriplegic
37:53
is hard enough, but taking care of a quadriplegic with
37:55
a traumatic brain injury is something that I would never
37:57
want my parents to do. So I made them agree.
38:00
to that. Yeah. Were you
38:02
nervous like the night before the
38:04
operation to install your neural length
38:06
device? What did you find yourself
38:08
thinking about as you went to
38:11
bed? Yeah, I wasn't nervous at all. I
38:13
was just excited. My buddy and
38:15
I were sitting around making jokes. I
38:18
don't know hanging out night before I just wanted to
38:20
get it over with honestly. What kind of jokes were
38:22
they making? We
38:24
were planning on releasing
38:27
some like cyborg related jokes, thinking
38:29
of things that only I would
38:31
be allowed to say. Just random
38:33
like turning to
38:35
phrases too like, oh, blew my mind picking
38:37
my brain, things like that. Yeah, just all
38:43
sorts. I'm curious if you
38:45
had a thought as you were heading
38:47
into this surgery of what the
38:49
first things you wanted to do would
38:52
be once you had a working brain
38:54
computer interface. Yeah, I
38:56
mean, I'm a big gamer. I wanted to play
38:58
games. That was one of the big things I
39:00
wanted to do. I also wanted to be able
39:02
to read. I mean, there's nothing like being able
39:04
to hold a book again, and the smell and
39:06
the feel of a paperback book. It's one of
39:09
my favorite things in the world and something that
39:11
I've missed for a long time. I can't do
39:13
that. And so
39:15
the next best thing is just being
39:17
able to read in general. I've
39:20
had to listen to audiobooks for the
39:22
last eight years, because I haven't been able to sit
39:25
in the same position to read a book. And
39:27
I had no way of turning pages. I can
39:29
do it sort of on a Kindle. But
39:32
I was listening to audiobooks and I don't
39:34
really like listening to audiobooks, to be honest.
39:37
Sometimes the narration is terrible. I don't
39:39
want to throw shade, but I remember
39:41
reading like an Aragon audiobook
39:43
and the voice that the narrator
39:46
had for the dragon made me turn it off
39:48
immediately. This is quite possibly the worst thing I've
39:50
ever heard in
39:54
my life. It was Gilbert Gottfried voicing the
39:56
dragon. No, that would have been amazing. I
39:58
would have listened to that. That on
40:00
repeat, honestly. No,
40:03
but just something as simple as that. Being able to
40:06
line my bed and read a book. There's
40:09
just something about it. So I was really
40:11
looking forward to that kind of stuff. So
40:13
how was the actual surgery? Was it long?
40:15
What was the recovery process like? Can you
40:17
tell us just about the actual implant? Yeah,
40:19
it was super, super quick. We
40:22
got to the hospital at five a.m. I think
40:24
I was scheduled for surgery at seven. There was
40:26
a lot of just getting me ready, getting me
40:28
in bed. The surgery was supposed to last between
40:31
four to six hours. They were
40:33
expecting there to be hang ups. They were expecting,
40:35
say, the needle on the robot to break.
40:38
They brought, I think, 20 iterations of that
40:40
needle in case it broke and then they
40:42
had to stop and replace it. And the
40:44
needle didn't break once. Just everything
40:46
performed above and beyond what they
40:48
expected. And so the surgery took
40:50
under two hours. And
40:52
then I was out
40:55
of surgery and they prescribed some pain
40:57
pills. I didn't take a single one. I don't
40:59
know, it was just so easy. The
41:01
worst part was I wasn't able to shower
41:03
for the first few days because my incision
41:06
needed to heal. But outside of that, the
41:08
recovery process was so easy. I didn't feel
41:10
any pain at all. Wow. And
41:13
when you woke up from your surgery, what
41:16
was actually different? Did you
41:18
feel different? And
41:20
then how long was it before you actually
41:22
got to turn on and use the Neuralink
41:24
device itself? Yeah, I mean,
41:27
I had a gnarly scar with some staples
41:29
in it. I mean, that was pretty freaking
41:31
sick. Like I was super happy about that.
41:33
I got some cool pictures. But
41:35
yeah, I think within, I don't
41:38
know, like an hour or two after my
41:40
surgery, they came in and connected me to
41:43
like a little tablet they had. What
41:45
do you mean connected you? Like literally plugged
41:47
something into your head? Or like what, is
41:49
it Bluetooth? It's a Bluetooth connection, yeah. So
41:51
they just wake up the implant with a
41:54
little coil, like a charging coil, almost like
41:56
the same thing that you put your phone
41:58
down on a mat to charge. It's
42:00
very similar. You just hold something over my head and it
42:02
wakes up. It's how you charge it as well
42:04
I put that in like a hat and I wear a hat and it
42:06
charges and
42:08
so they woke it up they Connected that
42:11
to a tablet And on
42:13
the screen they just showed a bunch of the
42:15
channels the channels are each electrode in
42:17
my brain and those electrodes are picking
42:19
up like neuron firing and
42:23
So they showed me say like eight channels and I
42:25
got to see like live like
42:27
real-time The neurons firing
42:29
in my brain and everyone just kind
42:31
of freaked out in the hospital room.
42:33
Everyone started cheering They were clapping which
42:35
was totally unnecessary It
42:38
was so awkward But yeah,
42:40
it was really really cool Right.
42:43
I'm so curious No And about the actual experience
42:45
of using this neuro length device because one of
42:47
the things that it allows you to do is
42:49
to control A cursor on a
42:51
screen as if you were like using
42:53
a mouse just by thinking but I've
42:55
never known like you know When I'm
42:57
using a computer, I'm not thinking I'm going
42:59
to place my cursor here. I'm going to
43:02
click this button I'm going I'm it's just
43:04
like it's a it's a much more kind
43:06
of fast twitch sort of unconscious response I'm
43:09
curious when you're trying to control a cursor
43:11
on a screen How intently do you have
43:13
to think about it in order for the
43:15
cursor to actually react? I
43:17
mean at first I wasn't very good
43:20
at it I was doing
43:22
what we call like attempted movements attempted
43:24
movements are basically, you know like
43:26
I said all the signals in my brain
43:28
are still firing so the threads are implanted in my motor
43:30
cortex and so when
43:32
I Attempt to move my hand those
43:35
signals are firing the implants picking that up
43:37
and an algorithm is Basically learning what I'm
43:40
trying to do and after doing it a
43:42
certain amount of times it'll Translate
43:44
that into cursor control in some way or
43:46
another and then it'll keep learning as
43:49
I was like a week in maybe two
43:51
weeks In I just thought to move the
43:53
cursor in one direction And
43:56
it moved it it Blew
43:58
my mind like it was it was wild.
44:01
But then over time, it
44:03
just becomes second nature. It's not like I'm
44:05
thinking like cursor come over here and I'm
44:07
waiting for it to get there or anything.
44:09
It's just it's very, very, like I said,
44:11
intuitive. And and how like fine grain
44:13
is the control you have over a cursor now? Is
44:16
it like roughly equivalent to like what it used to
44:18
be like when you were using a mouse or is
44:20
there still a gap there? I
44:22
would say it's very similar.
44:25
I'm not as quick with the cursor as a
44:27
lot of other people, but I don't think that
44:30
can't be made up. I think that
44:32
just comes with a bit more practice
44:34
and also just a bit more tweaking on
44:37
the software side. Like this is still very
44:39
early days, a few months in and we're
44:41
already where we're at, which is amazing. I
44:43
think by the end of my time
44:46
in this study, whenever that will be, I'll
44:48
be better than most people with a cursor.
44:50
Wow. I mean, one of the details that I
44:52
just love from the initial reporting
44:55
on your story is that after you
44:57
got your neural link, you played eight
44:59
straight hours of the video game Civilization
45:01
6. And I just love that because
45:04
it's like I imagine that,
45:06
you know, you had doctors years ago thinking that
45:09
in theory, if this ever, you know, this
45:11
brain computer interface thing ever worked, it would
45:13
allow people to, you know, do more types
45:15
of creative labor and and be more productive
45:17
at work. And you get it and just
45:20
immediately start gaming, which I think shows that you just
45:22
have the heart of a true gamer. Yeah,
45:24
yeah, I, I keep telling them that I
45:26
keep saying, you know, I'm just so unproductive
45:29
with this thing, like, by giving me more
45:31
things to do, like, I would much rather
45:33
be doing work. And they're like, no, just
45:35
do what you want to do. That's what
45:37
we want is to make you able to
45:40
play games to go surf the web and
45:42
do things that you want. It's not about,
45:44
you know, what other people believe you should
45:46
be doing or anything. It's just whatever makes
45:48
your life better. And waste your life like
45:51
the rest of us. I mean, it makes so
45:53
much sense, though, because it's like for for what,
46:00
eight years you had been deprived of being
46:02
able to just like use your hands to
46:04
play games. I love to play video games.
46:06
Like I play video games myself every week.
46:08
And I guarantee you that if I had
46:10
been in your shoes, I absolutely would have
46:12
been playing as civilization when they went to.
46:14
Yeah, yeah. Now, now is it also true
46:16
that you played Mario Kart? I
46:18
did. I did very early on, maybe a couple
46:20
weeks in, they hooked me up to a switch.
46:23
And that was very like hands on. They
46:26
were real time tweaking things right now we're
46:28
working on, you know, giving me that capability
46:30
on my own. So any day I
46:32
want I could just hop in. I
46:34
think that's pretty close. That's amazing. Now,
46:36
I want to know like how different
46:39
this is from other assistive technologies that
46:41
have come before it because we've had
46:43
things like eye tracking for computer control
46:45
before for people who have lost mobility.
46:47
So have you tried any other ways
46:49
of controlling computers before
46:52
this? And so how does this
46:54
implant stack up to other things that people have
46:56
been using to do similar things in
46:58
the past? Yeah, I've tried it all
47:00
from, you know, the first few
47:02
weeks, first few months I was in the
47:05
hospital after my accident, they had me trying
47:07
everything. And sure things have gotten better since
47:09
then, but they're just not even in the
47:11
same league as nearly eye trackers. They're
47:13
just not as good. A lot of
47:15
it has to do with, you know, being
47:18
centered on the screen, making sure that
47:20
your levels don't change. I
47:22
have really bad spasms. I'm very spastic. So
47:24
if I move at all, like my body
47:26
spasms to the right and I'm off center,
47:28
then the eye tracker doesn't really work anymore.
47:31
I've tried other things like a quad stick. And
47:33
there was a video of a guy, a quadriplegic who
47:36
was using a quad stick to play
47:38
things like Fortnite and stuff. I tried,
47:40
I tried playing like one
47:42
of the Call of Duty games come out like after
47:44
my accident. And I hopped on and tried the campaign.
47:46
And I think they're storming like the beaches of
47:48
Normandy. And I didn't even make it off the
47:50
beach. Unfortunately,
47:52
the case for so many people in that beach.
47:55
It's a very real, real case. That
47:59
happens to me too. play team shooters
48:01
because I just get my butt kicked
48:03
by 11-year-olds so I'm sympathetic.
48:05
What's your excuse, Rif? Nolan,
48:08
I want to ask you about thread
48:10
retraction because this is something that has
48:12
happened since your surgery. A lot of
48:14
the threads that connect the
48:16
Neuralink device to your brain actually
48:19
started retracting. I saw a figure
48:21
that 85% of them had retracted
48:23
and this was potentially endangering your
48:26
ability to use this device. Talk
48:29
with us about this and what the first signs
48:31
you experienced were that something wasn't right.
48:34
Yeah, a few weeks in, I just
48:36
started losing control of the cursor is what it
48:38
comes down to. It would start
48:40
drifting on me. I would want it to go
48:42
right and it would go left. I
48:45
could not get it to go down, things like that.
48:48
It just became impossible to use. About
48:51
a week later, so this was about three weeks
48:53
in, about a week later they told me that
48:56
they had seen some evidence of thread retraction but
48:58
I think they had only found out a day
49:00
before. They kept me in the loop the whole
49:02
time. They actually took a
49:04
scan of your brain and said it looks like we
49:06
can see that the threads have retracted? No, no. Brain
49:09
scans won't even show
49:11
the threads. What they can
49:13
do is look at the electrodes
49:17
over time and see
49:19
which electrodes on the threads
49:21
are sending signals and which ones are
49:23
sending strong signals or weak signals. They
49:26
can really tell which electrodes are
49:28
still in my brain. Right
49:30
now it's about 15% that are
49:32
still actively sending strong signals
49:34
in my brain. Do
49:37
they have any theories about why these threads have come
49:39
loose? Yeah, it has to do
49:41
with how much the human brain moves. Apparently,
49:43
they had thought that everything they
49:46
had read, all the surgeons they had
49:48
talked to, said that
49:50
the brain moves about one millimeter and
49:52
then when they implanted everything in my
49:54
brain, they found that my brain moves
49:56
actually three millimeters. So on a
49:58
scale of 3x what they read. expecting. So
50:02
obviously this is not like fixing a computer or
50:04
an iPhone or something where you can just like
50:06
open it up and fix it like the opening
50:09
up wouldn't involve opening up your head and your
50:11
skull and and doing brain surgery
50:13
on you again. So
50:15
how did they go about trying to fix this?
50:18
Yeah I offered them to go in take
50:21
out the implant and put in
50:23
a new one I was like if it's gonna
50:25
get me back to peak performance and that's what
50:27
I want if it's gonna help me stay in
50:29
the study I offered
50:31
that and they said no we're gonna
50:34
like stop take a step back and
50:37
see if we can fix this on
50:39
the software side which is ultimately what
50:41
ended up happening. They just tweaked the
50:43
way that they were recording signals from
50:46
the threads and from the electrodes
50:49
and that ended up working. There there were
50:51
a couple different ways they were recording the
50:54
neuron spikes in my brain. I mean
50:56
there's a ton of information coming from
50:58
the neurons at all times and they
51:00
were trying to interpret those
51:02
spikes in a certain way and they
51:05
had found that the way that we started was
51:07
the most efficient but then once all
51:09
the threads started retracting they needed to rethink that
51:11
and so they switched to a different way of
51:13
recording those signals and they found that that was
51:16
actually much better. And so
51:18
like where are you at now? Like
51:20
Neuralink has said that your performance with
51:22
the device is now better than it
51:24
was before all of this happened. Does
51:26
that continue to be the case? Yeah
51:28
yeah absolutely. I'm still getting
51:30
better too. I'm curious like
51:32
some of the people that I've talked to
51:35
in the tech community believe that BCIs are
51:37
going to be just a major mainstream technology
51:39
in the future not just for people with
51:41
disabilities but for basically anyone. I mean that
51:44
you know the next big platform shift may
51:46
not be people putting computers on their head
51:48
like with VR and maybe people putting computers
51:50
in their head and eventually we will all
51:53
be walking around with these brain implants. Based
51:56
on your experience do you think that is a
51:58
plausible future here? Yeah I don't see why. not.
52:01
I think they're safe. I think the
52:04
possibilities are endless with this technology. I
52:06
mean, we're just scratching the surface. I
52:09
don't know what kind of things people
52:11
are going to be able to
52:13
do with this in 10 years. I don't
52:15
think anyone really knows. There are
52:17
applications that we see can be
52:19
useful like helping cure paralysis or
52:22
different motor diseases, helping
52:24
cure blindness. But once we start getting all
52:26
of that, then that begs a
52:29
lot of other questions like, if we can do this, why
52:32
can't we do even more? And with the
52:34
AI revolution that we're in right now, how
52:37
can this be applied to all of these
52:39
pieces of hardware in our brain? I just
52:41
think we're in for an explosion,
52:43
like exponential growth in
52:46
this field, especially now that Neuralink has come
52:48
out with this. It's going to bring all
52:50
the other BCIs up and it's going
52:52
to push Neuralink to get even better. It's going
52:54
to be like a new space race, but in
52:56
the brain. You know,
53:00
I'm curious to get a sense of
53:03
what this has just been like for, you know,
53:05
you were talking earlier, I was really moved when
53:07
you were talking about just craving the experience of
53:09
holding a book in your hands again, something that
53:11
I take for granted. And now
53:13
presumably you've been able to read books,
53:15
you've been able to play games. What
53:18
is this done for you emotionally
53:20
to kind of get access to
53:22
some of those things that you had been missing
53:24
out on? I
53:27
mean, it's hard to even
53:29
put into words. Just
53:33
this amount of independence
53:35
that I've been given, it
53:38
changed my life for the
53:40
last few months. It's changed my
53:43
parents' lives. Little
53:46
things, I mean, very, very little things have
53:48
made huge differences. Like when I was able
53:50
to get a drink of water on my
53:52
own in the middle of the night because
53:54
I got like a little bottle
53:56
that stretched across my bed and allowed
53:58
me to have drinks. drinks in the
54:01
night, that relieved about 90%
54:04
of sleepless nights that my parents had.
54:07
And now with Nuralink, it's even
54:09
more than that. I'm able to do a
54:12
lot more on my own than I was never able
54:14
to do in the last eight years. I
54:16
don't have to wake anyone up in my family to
54:18
come help me in the middle of the night. I
54:20
don't have to feel guilty if at 2 a.m. I
54:23
want to connect and read or listen to an
54:25
audio book or play a game or just go
54:27
on and check my social media or text someone
54:30
back. I don't have to feel guilty about trying
54:32
to wake someone up. There's
54:35
just so much that I'm grateful for being able
54:37
to do this. And ultimately I want to use
54:39
it to help people, find
54:41
some way to help people, and I'm
54:43
on the path to doing that. And that's what
54:45
I've wanted since I was a kid, just to
54:48
find some way to help people. And after my
54:50
accident, I didn't think that was ever gonna happen.
54:52
I knew that I could still speak, but I
54:54
mean, who would want to listen to me speak
54:57
about nothing, I had no life experience to give
54:59
them. I guess now it's a bit different. But
55:02
yeah, I don't know. It's been such a
55:05
huge blessing to me, honestly. Yeah. I'm
55:08
curious, Nolan, what have your conversations with Elon
55:10
Musk been like? I
55:13
haven't had many. I talked to
55:15
him on FaceTime right before surgery. Was
55:18
like, hey, thanks for choosing me to
55:20
do this. I'm really excited,
55:22
really thankful, really blessed. And he
55:24
was like, yeah, this has been great. Really looking
55:26
forward to it, making a huge step. And
55:29
I said, let's rock and roll. He's like,
55:31
let's do it. And that was it. And
55:33
then after surgery, I spoke to him in
55:35
person. He came to the hospital. I
55:37
was still pretty drugged up on anesthesia, and
55:40
I couldn't get his sweet bomber jacket out
55:42
of my mind. I was just
55:44
lying in my bed thinking the whole time, don't
55:46
mention this bomber jacket. Don't mention the bomber jacket.
55:49
But it was cool. I think we have very
55:52
similar ideals about
55:54
what this can do for
55:56
humanity and where we
55:58
can go from here. And... just our drive
56:00
to help people in that way. I
56:03
think it's amazing that someone of
56:05
his caliber has stepped up and
56:07
stepped into this role
56:10
for helping people like me. I
56:12
mean, I never thought anything
56:15
like this would ever happen to me or
56:17
to people like me. And to have such
56:19
a high profile figure say, you know, I'll
56:22
take that on and I'll fight that. It's just
56:24
amazing. I'm curious what you
56:26
make of the promises that people like
56:28
Elon have made for how BCIs could
56:31
improve in our lifetimes. He did get
56:33
a little criticism a few years ago
56:35
for some statements he made at a
56:37
Neuralink presentation where he suggested that these
56:40
BCIs could eventually allow blind people to
56:42
see or give people with spinal cord
56:44
injuries, like the use of their full
56:47
bodies back. A lot of health experts
56:49
were very skeptical and they basically said,
56:51
it's irresponsible to say this stuff given that
56:53
the science is just nowhere near there yet. And
56:55
I'm curious how you feel when you hear the
56:57
kinds of lofty promises about what this technology may
56:59
be able to do someday. Do you get excited
57:01
or do you say like, hey, wait a minute,
57:04
let's like stick to what the science is capable
57:06
of now? No,
57:08
I'm super excited about it. It gives
57:10
me and people like me something to hope for.
57:13
Once you take away hope, that's the end
57:15
for most people. And for him
57:17
to promise something like that, even
57:20
if it never comes about, it's just the
57:22
fact that he's trying and he
57:24
sees it as a possibility. I
57:27
take that kind of passion to heart.
57:30
I don't agree with people who say that it's
57:33
irresponsible. I think it's a
57:35
reality from my perspective that it's going
57:37
to happen probably in my lifetime. And
57:39
if I'm irresponsible for saying that, then
57:41
like I'm sorry, but it
57:43
gives me something to look forward to and gives me
57:45
something to strive for and to work towards. And
57:48
maybe I fall short of that, but
57:50
I'll be damned if I don't give it my all. Yeah,
57:53
I mean, what's very clear to me about you, and on
57:55
is that you just have like a much
57:57
higher risk tolerance than I do. I
58:00
get nervous to go to the doctor to
58:02
get like some little you know thing and
58:04
you hear are saying I will volunteer I
58:06
will step up to be patient number one
58:09
for this potentially very severe use of technology
58:11
So yeah, my hats off to you for
58:13
just being willing to put your hand up
58:15
for it. Not for real. Yeah Thanks,
58:18
man. All right, no, well great
58:20
to talk to you. Thank you so much for your time.
58:22
Thanks so much for coming on. Thanks for having me guys.
58:24
I really appreciate it After
58:29
our interview we reached out to Neuralink to
58:31
confirm some of what Nolan shared with us
58:33
about his surgery But we didn't
58:35
hear back from them. You can read more about
58:38
his experience on their website When
58:40
I come back, we'll talk to my colleague Karen
58:43
Weiss about Microsoft's big AI enough Looking
58:55
for a new ride? Shop
58:58
for your next car from the comfort of
59:00
your home 100% online with Carvana Carvana
59:03
has a massive inventory including thousands
59:05
of cars under $20,000 So finding
59:07
a car that fits your budget
59:09
and lifestyle is hassle-free Carvana makes
59:11
car financing hassle-free Get pre-qualified
59:13
in minutes by answering a few quick questions Carvana
59:16
also offers customizable terms and payments as low
59:18
as zero dollars down because payment plans should
59:20
fit your plans Visit
59:23
carvana.com or download the app today to find
59:25
your next car without ever leaving the comfort
59:27
of home or anywhere else terms
59:29
and conditions may apply Well,
59:32
Casey, it is the most exciting time of
59:34
the year in the tech industry, which is
59:36
developer conference season That's right for a lot
59:38
of people Kevin this time of year is
59:40
about dads and grads for us. It's about
59:42
API's So
59:44
last week we talked about Google's I
59:46
owe developer conference and everything they showed
59:48
off and this week Microsoft had its
59:51
big annual developer conference called build you
59:53
did not go in person. Did you
59:55
I did it and candidly? Well, while
59:57
I read some coverage of this I
59:59
want to learn so much more because
1:00:01
I only had so much time left
1:00:03
over after I finished researching the filmography
1:00:05
of Scarlett Johansson. Right. So Microsoft obviously
1:00:07
is also very excited about AI. They
1:00:09
have been building out a lot of
1:00:11
their own AI tools and products and
1:00:13
services. And this week at
1:00:16
Build, they actually demoed some new hardware
1:00:18
that they are making that is sort of built
1:00:20
around AI. Now, isn't putting AI
1:00:23
directly into the computer's house Skynet began
1:00:25
the Terminator films? I'm not sure.
1:00:28
It's been a while since I watched those movies. I
1:00:30
think Microsoft gets less coverage by a tech journal.
1:00:32
It's like you and me then it deserves in
1:00:34
part because a lot of what they do is
1:00:36
like boring enterprise software stuff. But they
1:00:38
are the biggest company in the world and
1:00:40
they have been investing in AI significantly over
1:00:42
the past few years. And I would say
1:00:44
between their stake in open AI and all
1:00:46
of their own AI projects, they're just a
1:00:48
major, major player in this world. Yeah, it
1:00:51
is true. I probably don't pay as much
1:00:53
attention to Microsoft as I should. And it
1:00:55
is for a somewhat selfish reason, which is
1:00:57
I just use Macs. And so
1:00:59
sometimes it feels like this stuff just is not as
1:01:01
relevant to my life. But let's face it, for
1:01:03
most of the working world, they are doing their
1:01:06
work on a PC. And so if Microsoft says
1:01:08
we're putting AI in it, then we should be
1:01:10
paying attention. Yeah. So to
1:01:12
talk about this, we're going to bring on my
1:01:14
colleague Karen Weiss, who covers Microsoft for the New
1:01:16
York Times. She went to Build and she's going
1:01:18
to tell us all about what Microsoft announced. Karen
1:01:23
Weiss, welcome
1:01:26
to Hard Fork. Happy
1:01:28
to join you guys. So
1:01:32
Karen, you actually went up to Microsoft's
1:01:34
headquarters earlier this week for their annual
1:01:36
Build conference. So just set the scene
1:01:38
for us a little bit. Like what
1:01:40
was it like? How did it compare
1:01:43
to previous experiences? Well, I
1:01:45
think what was unique was on Monday,
1:01:47
they tried to really hype this AI
1:01:49
PC announcement and they did
1:01:51
something on the campus like the big headquarters
1:01:54
in Redmond just outside of Seattle. And
1:01:56
you had to be in person there. They weren't live streaming it
1:01:58
and Satya was going to get it. to give a keynote
1:02:00
at it. So it was trying to definitely build
1:02:03
attention. And so, you know, they tried, it was like a lot
1:02:05
of hoopla. There's a lot of music
1:02:07
in the background and stuff like that.
1:02:09
And they tried to recreate the magic,
1:02:12
if you will, when they launched the
1:02:14
Bing, Sydney chatbot that was
1:02:16
not spoken of about a year and whatever
1:02:18
ago. Yeah, that all
1:02:21
turned out great. So, Karen, one
1:02:23
of the things that Microsoft announced are
1:02:25
these things called Co-Pilot Plus PCs, which
1:02:28
as I understand it, are basically a personal
1:02:31
computer, a Windows PC that is basically
1:02:33
built to run AI and
1:02:36
that runs it very fast and that is
1:02:38
sort of all wrapped around the
1:02:40
capabilities of these AI models. So what did
1:02:42
they actually announce and sort of how is
1:02:44
it different from the Windows
1:02:46
PC that people use today?
1:02:49
Yeah, the main thing is that these
1:02:51
PCs have a bunch of AI
1:02:53
models, AI systems locally on the
1:02:55
computer. And they can run
1:02:57
different AI tools or
1:03:00
models because they have this
1:03:02
new type of processor, essentially. It's called
1:03:04
a, bear with me, an NPU, a
1:03:06
Neuro Processing Unit. And it
1:03:08
is very quick with really little drain on
1:03:10
your battery because the problem you have is
1:03:12
when you run AI, it's like super
1:03:14
intensive, right? It's running all these calculations
1:03:17
constantly. And so this is a whole
1:03:19
new generation of chip. Right. So
1:03:21
I guess I'm just struggling to understand what it
1:03:24
means for AI to be able to run
1:03:26
locally because I also have run large
1:03:28
language models locally on my hard drive
1:03:30
before. And is that what
1:03:32
they're saying? Is that if you wanna
1:03:34
run something like a chatbot, it'll just
1:03:36
be much faster to do or are
1:03:38
there actually capabilities that these will have?
1:03:41
Is it gonna be so deeply woven
1:03:43
into the operating system that people's
1:03:45
experience of their computer will actually change
1:03:47
somehow? Right, so they are hoping
1:03:49
the latter. You'll be shocked to hear. But
1:03:51
basically the idea is because there are these
1:03:53
things that are run locally, they can have
1:03:55
access to information that is only stored locally
1:03:58
and they can be faster and more. interactive
1:04:00
in a way where, you know, even right now, when
1:04:02
you go into a chatbot, you ask your question, right?
1:04:05
And it like takes a second and it goes boop, boop, boop,
1:04:07
boop, boop, and each little word kind of comes out one at
1:04:09
a time. This would speed that all
1:04:11
up as well. And so they're hoping that by putting
1:04:13
these different, they have said there were more than 40
1:04:15
models that can run, can kind of
1:04:17
come pre-installed, so to speak, on
1:04:19
the laptops. And their hope is that
1:04:22
developers now start playing with those and build tools
1:04:24
off of it. And they
1:04:26
liken it to the iPhone when the iPhone
1:04:28
said, oh, wait, here's a, you know, GPS,
1:04:30
here's a altitude meter, like whatever
1:04:33
kind of tools are in the hardware, people start
1:04:35
building it and then you get Uber
1:04:37
or whatever type of system that then made the
1:04:39
iPhone this kind of like key platform for people
1:04:41
to build off of. And so
1:04:43
I think there's a reason why they announced this before
1:04:45
their big event for developers is because they're trying to
1:04:47
say to developers, build for this, there's a
1:04:50
future here. Now I think
1:04:52
speed matters a lot. I do think that
1:04:54
that can really change the way that this
1:04:56
stuff gets used. And I can absolutely see
1:04:58
this becoming a developer platform. But
1:05:00
at the same time, Karen, I am reminded of
1:05:03
the last time that Kevin and I went up
1:05:05
to Redmond for an event. And we
1:05:07
were told Bing has AI now and AI
1:05:09
is going to change everything. And
1:05:11
I think we were optimistic that maybe that would
1:05:14
be the case. And then you fast forward to
1:05:16
today and Google is still by far the most
1:05:18
dominant search engine in the market. So I wonder
1:05:20
as you're hearing this presentation telling
1:05:22
us that AI chips inside PCs
1:05:25
are going to be a next
1:05:27
generation kind of PC that is truly going
1:05:30
to change everything. I wonder, what
1:05:32
did you think of that? I think
1:05:34
the difference between this and Bing
1:05:36
is Windows. So Windows is this
1:05:39
ubiquitous operating system that Microsoft controls.
1:05:41
So when they announced this, it
1:05:43
wasn't just them trying to get up
1:05:45
move from a tiny market share. They
1:05:47
have this enormous market share. And they had all
1:05:49
the biggest laptop makers in the world there showing
1:05:52
off versions of these devices. So I think they
1:05:54
have a power or an influence on the PC
1:05:56
in a way that they don't necessarily in search
1:05:58
or with Bing. That said, they
1:06:00
have to prove the utility of these things. And
1:06:02
part of why running the models locally
1:06:04
is important is it's cheap, right? You pay for
1:06:07
the processor upfront when you buy the laptop. But
1:06:09
every time you ping it, you're not spending a
1:06:11
penny. Whereas if you're inferencing things in the cloud,
1:06:13
if you're running things off the cloud, it
1:06:15
becomes really expensive for a developer to think, oh, can I
1:06:17
even afford to offer this product
1:06:20
to people? Now, if they can just do it locally, it might
1:06:22
not be quite as good. Like the language
1:06:24
models that can run on a laptop now are not
1:06:26
as good as GPT-4. But they
1:06:28
say about as good as GPT-3.5,
1:06:31
which was what Caffe d'Ebuty was initially
1:06:33
launched with. So you can
1:06:35
get some good enough uses is the theory.
1:06:38
But again, they have to find the kind
1:06:40
of key use cases that will show it.
1:06:42
I was a little surprised. The examples they
1:06:44
released with were not... They
1:06:47
were clearly trying to demonstrate certain capabilities.
1:06:49
Like what? What did they
1:06:51
demonstrate? Some were a little more
1:06:53
speculative and futuristic, saying this is the kind
1:06:55
of thing you would be able to do.
1:06:57
So like one was a dad
1:07:00
speaking with a voice
1:07:02
chatbot, essentially, asking for help on how
1:07:04
to solve the particular Minecraft situation. And
1:07:06
the voice thing, you can come in
1:07:08
and then say, oh, you
1:07:11
need these materials to build this new thing
1:07:13
in Minecraft. And therefore, oh, hurry, this thing
1:07:15
just popped up. Go run over there and
1:07:17
hide in this basement thing. And
1:07:19
because gaming is real time, you
1:07:21
can't do that if you're pinging in the cloud. And
1:07:24
so this was an example of a way where you could
1:07:26
bring in general AI and assistance into
1:07:28
like a real live moment, essentially.
1:07:30
But that was one of the
1:07:32
more futuristic products. It's like not launching
1:07:34
with that. Right. I
1:07:36
want to ask about this other feature
1:07:38
that they announced that I think got
1:07:41
a lot of attention, which is something
1:07:43
called recall. What is
1:07:45
recall? They
1:07:47
liken it to a photographic
1:07:49
memory. And it's actually
1:07:51
kind of makes sense because what it does
1:07:53
is it basically builds a history of everything
1:07:55
you have looked at on your laptop. And
1:07:58
it literally takes constant screenshots of your
1:08:00
screen stores those locally and
1:08:02
then you can ask it, oh
1:08:04
where was that thing that Kevin sent me that
1:08:06
was like kind of weird and he laughed when
1:08:09
he told me about it. And
1:08:11
since you and I are of course chatting on teens Kevin,
1:08:13
on my podcast, it would be like, oh
1:08:15
that we can pull that up here because the
1:08:17
transcript says that he laughed in this moment and
1:08:20
you can kind of scrub back and forth
1:08:22
in time trying to visually
1:08:24
look for what you were searching
1:08:27
for. So basically if you've always wondered what
1:08:29
would it like to have an FBI agent
1:08:31
living inside your computer, you can now have
1:08:33
that. Perfect. So
1:08:36
there's an app called Rewind that has been doing
1:08:39
something that sounds very similar to this,
1:08:41
but just so I'm clear, this is
1:08:43
taking screenshots, it is storing them
1:08:45
on your machine and then it is allowing
1:08:48
you to use generative AI to sort of
1:08:50
search back through your previous encounters
1:08:53
with your computer and say what was that
1:08:55
restaurant menu I was looking at last Tuesday
1:08:57
or whatever. But I'm curious, Karen, what you
1:08:59
think the target audience for this, who is
1:09:01
actually going to use this
1:09:04
feature and what examples did Microsoft give of how
1:09:06
it might be useful? They gave
1:09:08
some of kind of personal uses. They
1:09:10
had this very funny example of this
1:09:12
woman trying to get a dress for
1:09:14
her grandma and they talked about how
1:09:16
she had searched things online, she had
1:09:18
chatted in Discord with her abuela which
1:09:20
I thought was very funny because my
1:09:23
grandma used to AOL instant message me but I'm not sure
1:09:25
how many grandmas are on Discord. But
1:09:27
the idea was that there's this digital record
1:09:29
and you can go back and be like,
1:09:31
oh, what was that sparkly one? And
1:09:34
you can use language like that and because
1:09:36
it has visual intelligence, it can
1:09:38
go back and look for the sparkly blue dress you
1:09:41
had looked for and then, oh, didn't abuela say that
1:09:43
she actually really liked a pantsuit instead
1:09:45
and then you can kind of find the pantsuit. So
1:09:47
that's kind of a personal one. I was recently shopping
1:09:49
for jeans and part of me is like, oh, I
1:09:51
could see that. There was the jeans that I liked,
1:09:53
but also I'm scared if I were to type in
1:09:55
like, like jeans I was just looking
1:09:57
at. How many white jeans would show up?
1:10:00
It's kind of mortifying to think about how many wide-legs
1:10:02
you need to find to find a good pair. Totally.
1:10:05
So, you know, I can see all those
1:10:07
use cases. At the same time, all of
1:10:09
us are journalists. We often talk to people
1:10:11
off the record confidentially. We have sensitive information
1:10:13
on our laptops. We're not alone in that.
1:10:15
I think most people working jobs
1:10:18
have some sort of confidential information that is
1:10:20
on their computer. So I hear everything that
1:10:22
you've just described, Karen, and I think, absolutely
1:10:24
not. I tried Rewind for a while. I
1:10:26
found it terrifying and deleted it from my
1:10:29
phone. What is your thinking right
1:10:31
now about whether a lot of people are
1:10:33
going to be willing to invite this level
1:10:35
of surveillance onto their devices? Yeah, I mean,
1:10:37
the default is definitely to
1:10:39
take all of it. You can go back in and
1:10:42
manually delete certain days. You can have
1:10:44
it opt out of certain applications. But
1:10:46
like if you're going to opt out of the
1:10:48
web search, like that's one of the main uses
1:10:50
for it. So it's a big question. That's that
1:10:52
it's defaulting to being there. And one of the
1:10:55
things when I was kind of researching before the
1:10:57
event was a lot of people
1:10:59
haven't interacted with chat bots that haven't used
1:11:01
chat CPT, but they get a basic Windows
1:11:03
laptop from their work. And there's a big
1:11:05
old button that says co-pilot on it. And
1:11:07
these tools will be there. And for many
1:11:10
people, it might be the first time that
1:11:12
they really have exposure to it. Also, I'm
1:11:14
not sure people will understand the technology behind
1:11:16
it. And that is literally taking a picture
1:11:18
of everything that you're you're doing. And we should
1:11:20
say, like, you know, Microsoft
1:11:23
has said that all of these screenshots
1:11:25
stay on the device itself. They are
1:11:27
not being sent to Microsoft. Microsoft has
1:11:29
also said, you know, they're not doing
1:11:32
any kind of like content moderation on
1:11:34
them. So like, if you've been looking
1:11:36
at your bank account information on your PC, they're not
1:11:38
gonna like scrub that from the screenshots, but that it
1:11:41
will all stay locally on the device. And so only
1:11:43
the person whose device that is can access that. But
1:11:45
I think there are a lot of questions that
1:11:47
people will have, especially, you know,
1:11:49
if this is maybe a corporate
1:11:51
issued computer, does my employer then
1:11:53
have the ability to go back and look at
1:11:56
screenshots of every time I've ever used this computer?
1:11:58
Of course they will. gets normalized,
1:12:00
I can imagine employers handing
1:12:02
you your shiny new co-pilot
1:12:05
plus PC and saying,
1:12:07
you have to leave this recall feature
1:12:09
on. And then if we ever have
1:12:11
a disciplinary issue with you or we
1:12:14
are just suspicious about you for whatever
1:12:16
reason, we can review everything you have
1:12:18
ever done on your corporate issued laptop.
1:12:20
That seems like a nightmare dystopia to
1:12:23
me. And they can search for it
1:12:25
easily. Yeah. Right. Your
1:12:29
only chance to survive that is just
1:12:31
that they're using Bing search, which doesn't work
1:12:33
for the time. Well, Karen, that brings me
1:12:36
to one of my other questions about all
1:12:38
of this stuff that Microsoft announced this week.
1:12:40
It's like, who is it for? Because I
1:12:43
think traditionally Microsoft, a big part of their
1:12:45
business is selling to businesses, its enterprise customers,
1:12:47
its large companies that already run a lot
1:12:49
of Windows PCs that already use Outlook and
1:12:52
Teams and all the other Microsoft products. And
1:12:54
they can just kind of keep adding to
1:12:56
that bundle. So do you see these AI
1:12:58
PCs and all the features that are on
1:13:01
them as being aimed at businesses or are they
1:13:03
really making a consumer play here? You know,
1:13:05
I initially thought beforehand that it would
1:13:07
be more of the businesses, businesses by
1:13:10
literally the majority of laptops now. When was
1:13:12
the last time you guys refreshed your PC?
1:13:14
Mine was in 2011, I believe. You know,
1:13:17
you're eligible every three years at the New York Times.
1:13:19
Oh no, no, my personal one. When was the last
1:13:21
time you bought a personal one? Your PC is 13
1:13:23
years old, Karen. What are you using for? Are you
1:13:26
running Windows 98? XP? This is the MacBook and it's
1:13:28
lasted and like for
1:13:32
the internet, right? Everything's on the internet that
1:13:34
I do now. I'm not like ripping my
1:13:36
CDs anymore and taking my music from one
1:13:38
thing to another. You're not mining Bitcoin? I'm
1:13:41
not. I was surprised though. They
1:13:44
definitely sort of benchmarked it and kind
1:13:46
of compared it to the MacBook Airs. So
1:13:48
they're smaller, they are lighter,
1:13:50
they say they start at $1,000. So
1:13:53
they're not like Chromebook prices, but they're
1:13:55
also not a MacBook Pro super
1:13:58
heavy thing. Karen,
1:14:00
I'm curious how this positions
1:14:02
Microsoft against Apple. Apple's having
1:14:04
its own developer conference next
1:14:06
month, and there's a
1:14:08
lot of speculation that Microsoft's announcements at
1:14:11
this event this week were designed to
1:14:13
take some attention away from anything Apple might
1:14:16
announce in just a few weeks. Apple is
1:14:18
also expected to do a lot around generative
1:14:20
AI in its own hardware
1:14:22
products. So what is the competition between
1:14:25
Microsoft and Apple like right now? These
1:14:28
chips, these new NPUs, do
1:14:31
propel the performance of these laptops
1:14:35
basically in the realm of Apple now.
1:14:37
So it's in the same class because
1:14:39
they've changed over this tip architecture
1:14:42
or style essentially now to
1:14:44
a model that actually Apple had been pursuing. So
1:14:47
the question is, will people make the jump from
1:14:49
one ecosystem to another? Will their
1:14:51
employers make a jump from one ecosystem to
1:14:53
another? All of that stuff is I think
1:14:56
I don't know the answer to that. But they are clearly
1:14:58
trying to show that they are
1:15:00
building even their hardware with this
1:15:02
AI first mentality. And again,
1:15:04
if people want that, if they can
1:15:06
demonstrate the utility of that, that's the
1:15:08
question about all of this. Yeah,
1:15:11
I just think so much of this
1:15:13
is going to depend on how Microsoft
1:15:16
implements all this stuff and honestly how
1:15:18
annoying they are about it. So
1:15:20
I have a Windows PC. It's my daily
1:15:22
sort of PC that I use. And
1:15:25
I would say most of the time I really like
1:15:27
it, but then there are just these times when
1:15:29
it's very clear that Microsoft is just getting a
1:15:31
little bit greedy and they just start popping things
1:15:33
up or putting things in weird places. Like the
1:15:35
other day, I was using my computer and I
1:15:38
got a Skype news alert. Have you seen these
1:15:40
yet? I've seen somebody
1:15:42
posting these on threads. And it's so funny.
1:15:44
It's so funny. I was just minding my
1:15:46
own business, doing email. Up comes this little
1:15:49
notification that says, Skype says the
1:15:51
US economy added this many jobs last month. And
1:15:53
I'm like, why is Skype talking to me? Did
1:15:55
I ask Skype to talk to me? No, I
1:15:57
did not. But that is just a classic case of
1:15:59
Microsoft. So trying to juice engagement by doing something
1:16:01
that I think a lot of people would feel
1:16:03
is very annoying. So I can see these AI PCs
1:16:06
being very useful and I want to try one, but
1:16:08
I think if they can't resist sort of, I don't
1:16:10
know, just trying to nudge you into using it
1:16:12
more and more or in different ways, I think
1:16:14
that is going to turn a lot of people off.
1:16:16
I mean, I think a lot of AI things
1:16:18
are going to be nudging you more and more
1:16:20
because they want discoverability is that phrase that you
1:16:22
hear a lot about about the features that these
1:16:24
things can do. You know, this is a huge
1:16:26
Alexa problem, right? People know Alexa can
1:16:28
do the timer or whatever, but she can
1:16:32
do more and they want you to do more,
1:16:34
but like they got to push it to you.
1:16:36
Otherwise, how do you know? So I think that's
1:16:38
a very hard urge to resist, even though I
1:16:40
completely understand and agree. Yeah, I mean, Kevin, I'm
1:16:42
like you, I would like to give one of
1:16:44
these things a try, see what it can do,
1:16:46
see how fast this AI is once you get
1:16:48
it on the device. But I do continue to
1:16:50
have trust issues with Microsoft. This is a company
1:16:53
that just last month started testing ads in the
1:16:55
Windows 11 start menu. So every
1:16:57
time you go to like, look at the
1:16:59
programs on your computer now, you might just
1:17:01
have to see an ad. And I
1:17:03
don't know the more AI is on my computer,
1:17:05
the less I'm excited about a company that is
1:17:07
looking to shut ads into different parts of the
1:17:09
interface. You know, imagine you're you're going to, you
1:17:11
know, shop for jeans for Abuela
1:17:13
and then, oh, you're looking for that. That
1:17:15
well, here's an ad for that now, Karen.
1:17:18
Right. So there's just a lot of stuff
1:17:20
in here that sort of has my eyebrows
1:17:22
arsed. Karen, did you
1:17:24
actually get to try any of this stuff? Did you get
1:17:26
to get your hands on one of their new AI PCs?
1:17:28
Yeah, they had like a demo station set
1:17:31
up. And so like, it's populated with like
1:17:33
all the demo data. And so the recall
1:17:35
example was hard to know what it would
1:17:37
be like when you had a real body
1:17:40
of your own data. Would it feel super
1:17:42
creepy? Would it be really useful? Would it
1:17:44
be not useful because it returns so much information
1:17:46
that you can't actually scrub through it all? I
1:17:48
understand the problem they're trying to solve with that.
1:17:51
Like, I think we all have the zillion tabs
1:17:53
open that you keep open just so
1:17:55
you don't forget about it. So I understand that
1:17:57
like impetus behind it, but it was hard to get a
1:17:59
sense of... for that for me of
1:18:01
like would I personally use this and
1:18:04
like it essentially because it was just filled
1:18:06
with all this dummy content essentially. Right.
1:18:08
You know, you know, Karen dummy content was actually
1:18:10
the original title for this podcast. Some
1:18:14
people think it's still could be. Nailed
1:18:17
it. All right, Karen Wise, thanks for
1:18:19
coming. Thanks guys. Thank
1:18:31
you. Looking
1:18:39
for a new ride? Check for your
1:18:41
next car from the comfort of your
1:18:43
home. One hundred percent online with Carvana.
1:18:45
Carvana has a massive inventory, including thousands
1:18:47
of cars under twenty thousand dollars. So
1:18:49
finding a car that fits your budget
1:18:51
and lifestyle is hassle free. Carvana makes
1:18:53
car financing hassle free. Get pre-qualified in
1:18:55
minutes by answering a few quick questions.
1:18:57
Carvana also offers customizable terms and payments
1:18:59
as low as zero dollars down because
1:19:01
payment plans should fit your plans. Visit
1:19:03
carvana.com or download the app today to
1:19:05
find your next car without ever leaving
1:19:07
the comfort of home or anywhere else.
1:19:10
Terms and conditions may apply. Hard
1:19:14
Fork is produced by Whitney Jones and Rachel
1:19:16
Cohn. We're edited by Jen Point. We're fact
1:19:18
checked by Caitlin Love. Today's show
1:19:20
was engineered by Chris Wood. Original
1:19:23
music by Marion Lozano, Rowan Nemestow
1:19:25
and Dan Powell. Our
1:19:27
audience editor is Nel Gologli. Video
1:19:29
production by Ryan Manning and Dylan Bergison. If
1:19:32
you haven't checked out our YouTube channel, you
1:19:34
can find it at youtube.com/hard fork. Special
1:19:37
thanks to Paula Schumann, Kui-Wing Tam,
1:19:39
Kate Laprasi and Jeff Rameranda. As
1:19:41
always you can email us at hard fork at any
1:19:44
time. What's
1:20:14
up sandwich heads? Today on Steve-O's Sandwich Reviews we've
1:20:16
got the tips and tricks to the best sandwich
1:20:18
order and it all starts with this little guy
1:20:20
right here, Pepsi Zero Sugar. Marshall's
1:20:22
a pastrami craving a cubano. Yeah, sounds
1:20:24
delicious but boom! Add the crisp, refreshing
1:20:26
taste of Pepsi Zero Sugar and cue
1:20:28
the fireworks. Lunch, dinner or late night
1:20:30
it'll be a sandwich worth celebrating. Trust
1:20:32
me, your boy's eaten a lot of
1:20:35
sandwiches in his day and the one
1:20:37
thing I can say with absolute fact,
1:20:40
every bite is better with Pepsi.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More