WEBVTT
00:00:00.000 --> 00:00:02.580
This is the Read Multiplex podcast, and this
00:00:02.580 --> 00:00:05.440
is a special analysis of a company by Apple that
00:00:05.440 --> 00:00:08.279
only Brian Rommel could give. If you like his
00:00:08.279 --> 00:00:10.320
work or get anything out of it, become a member
00:00:10.320 --> 00:00:12.619
or at least buy him a coffee. Absolutely. Yeah.
00:00:12.740 --> 00:00:16.199
So with that out of the way, I mean, what if
00:00:16.199 --> 00:00:18.940
I told you Apple just spent up to two billion
00:00:18.940 --> 00:00:21.039
dollars to make sure Siri can read your mind
00:00:21.039 --> 00:00:23.320
before you even open your mouth? It sounds totally
00:00:23.320 --> 00:00:25.359
insane when you say it like that out loud. Right.
00:00:25.539 --> 00:00:28.760
It sounds like pure sci -fi. But today we're
00:00:28.760 --> 00:00:31.280
going into some incredibly strange, fascinating
00:00:31.280 --> 00:00:35.219
territory because, well, we're so used to the
00:00:35.219 --> 00:00:37.880
idea that action in our physical world is like
00:00:37.880 --> 00:00:40.880
thunder. Yeah, the thunder analogy is perfect
00:00:40.880 --> 00:00:42.840
for this. You see a lightning strike, you wait
00:00:42.840 --> 00:00:44.880
a beat, and then you hear the boom. Cause and
00:00:44.880 --> 00:00:47.439
effect, right? Sound and action. You speak a
00:00:47.439 --> 00:00:49.240
command to your phone, it lights up, it does
00:00:49.240 --> 00:00:51.479
the thing. Because we live in the thunder. I
00:00:51.479 --> 00:00:53.200
mean, we've built literally all of our voice
00:00:53.200 --> 00:00:56.039
technology around waiting for that acoustic boom
00:00:56.039 --> 00:00:59.079
to hit a microphone. Exactly. But what if a computer
00:00:59.079 --> 00:01:01.119
just didn't need to wait for the thunder anymore?
00:01:01.320 --> 00:01:03.179
What if it could see the invisible lightning
00:01:03.179 --> 00:01:05.879
happening like right under your skin milliseconds
00:01:05.879 --> 00:01:09.000
before you actually make a sound? Okay, let's
00:01:09.000 --> 00:01:11.519
unpack this. Let's do it. We are looking at a
00:01:11.519 --> 00:01:14.560
massive, completely under -the -radar acquisition
00:01:14.560 --> 00:01:18.799
by Apple that went down in January 2026. They
00:01:18.799 --> 00:01:21.480
bought a company called Q .AI, formerly known
00:01:21.480 --> 00:01:26.340
as Q, or QLTD, for a reported $1 .5 to $2 billion.
00:01:26.840 --> 00:01:29.299
Which is just a staggering amount of money. Right.
00:01:29.359 --> 00:01:31.500
That makes it Apple's second largest acquisition
00:01:31.500 --> 00:01:34.420
in history, right behind Beats. And we're using
00:01:34.420 --> 00:01:37.000
Brian Rommel's exclusive analysis today to look
00:01:37.000 --> 00:01:39.379
past the mainstream media's incredibly superficial
00:01:39.379 --> 00:01:41.269
coverage. Ridge of the Steel. They really missed
00:01:41.269 --> 00:01:43.010
the mark on this one. I mean, completely. Oh,
00:01:43.049 --> 00:01:45.230
entirely. So our mission for you on this deep
00:01:45.230 --> 00:01:47.549
dive is to understand how this purchase isn't
00:01:47.549 --> 00:01:50.069
just an audio upgrade. You know, it represents
00:01:50.069 --> 00:01:51.989
a foundational leap towards something called
00:01:51.989 --> 00:01:54.450
connectionless intent reading. And we really
00:01:54.450 --> 00:01:56.769
need to emphasize the sheer scale of that number.
00:01:57.049 --> 00:02:00.739
Two. billion. That is not an acqui -hire. No,
00:02:00.799 --> 00:02:03.760
definitely not. That is not buying a neat little
00:02:03.760 --> 00:02:07.159
feature to patch a bug in iOS. It is a foundational
00:02:07.159 --> 00:02:10.620
bet on a completely new paradigm for human -computer
00:02:10.620 --> 00:02:14.500
interaction. It's bypassing the acoustic requirement
00:02:14.500 --> 00:02:17.439
of computing entirely. Right. But if you look
00:02:17.439 --> 00:02:19.699
at the public narrative right now, like there's
00:02:19.699 --> 00:02:22.180
this tech news carousel that's been blowing up
00:02:22.180 --> 00:02:25.539
on Instagram all week. The whole Q .AI purchase
00:02:25.539 --> 00:02:28.699
is being framed as a routine AI shopping trip.
00:02:28.819 --> 00:02:31.360
Oh, yeah. The whispered speech. Yeah. They're
00:02:31.360 --> 00:02:34.020
saying, oh, Apple spent a couple billion dollars
00:02:34.020 --> 00:02:37.699
for whispered speech technology. or noise -robust
00:02:37.699 --> 00:02:40.759
audio. The narrative is basically, well, Apple
00:02:40.759 --> 00:02:42.800
bought this, so Siri finally works when you're
00:02:42.800 --> 00:02:44.800
standing on a noisy subway platform. Which is
00:02:44.800 --> 00:02:47.659
hilarious. Even Johnny Sruji, you know, Apple's
00:02:47.659 --> 00:02:50.120
hardware chief, publicly called the Q .AI firm
00:02:50.120 --> 00:02:53.199
remarkable in this really vague, noncommittal
00:02:53.199 --> 00:02:55.360
way. Which is exactly what Apple wants everyone
00:02:55.360 --> 00:02:57.560
to think. Whispered speech is honestly the perfect
00:02:57.560 --> 00:02:59.419
cover story. Because it sounds boring. Right.
00:02:59.439 --> 00:03:01.780
It's a plausible, easily digestible narrative
00:03:01.780 --> 00:03:03.639
that they can say out loud without tipping their
00:03:03.639 --> 00:03:05.780
hand to competitors. But, I mean, think about
00:03:05.780 --> 00:03:07.780
it. out the economics for a second. Solving background
00:03:07.780 --> 00:03:10.840
noise on a subway is a $50 million software problem.
00:03:11.379 --> 00:03:13.960
You don't drop $2 billion to build a better noise
00:03:13.960 --> 00:03:16.840
gate. That's what I'm saying. Framing a $2 billion
00:03:16.840 --> 00:03:19.620
acquisition as a noise cancellation upgrade is
00:03:19.620 --> 00:03:22.699
like, I don't know, buying a billion dollar supercomputer
00:03:22.699 --> 00:03:25.000
just so you can use the calculator app to tip
00:03:25.000 --> 00:03:26.900
your waiter. Yeah, makes zero sense. It makes
00:03:26.900 --> 00:03:29.330
no sense at all. So if it's not a chatbot. And
00:03:29.330 --> 00:03:31.469
it's not noise cancellation. What did Apple actually
00:03:31.469 --> 00:03:34.330
buy? They bought a way to read words that are
00:03:34.330 --> 00:03:36.710
not fully spoken. Which honestly gives me chills
00:03:36.710 --> 00:03:39.710
just thinking about it. But to really grasp how
00:03:39.710 --> 00:03:42.430
you read an unspoken word, we can't just look
00:03:42.430 --> 00:03:45.930
at January 2026. I want to take a step back.
00:03:46.169 --> 00:03:48.389
We have to look at the history. Yeah, we actually
00:03:48.389 --> 00:03:50.830
have to travel back to trace a 13 -year circle
00:03:50.830 --> 00:03:54.830
that starts in 2013 with one very specific founder.
00:03:55.150 --> 00:03:57.750
Eviad Maisels. That's the guy. And this history
00:03:57.750 --> 00:04:00.580
isn't just trivia. You know, it is the absolute
00:04:00.580 --> 00:04:03.319
blueprint for what is happening today. Without
00:04:03.319 --> 00:04:05.879
understanding his track record, the Q .I deal
00:04:05.879 --> 00:04:08.500
looks like magic. But with his track record,
00:04:08.680 --> 00:04:11.039
it looks like inevitable physics. Let's lay out
00:04:11.039 --> 00:04:13.659
that track record for everyone. Back in November
00:04:13.659 --> 00:04:17.740
2013, Apple bought a Tel Aviv -based company
00:04:17.740 --> 00:04:22.480
called PrimeSense for about $360 million. A pretty
00:04:22.480 --> 00:04:24.779
big deal at the time. Yeah. Now, if you were
00:04:24.779 --> 00:04:27.029
paying attention to tech back then, You probably
00:04:27.029 --> 00:04:29.269
know PrimeSense as the company that built the
00:04:29.269 --> 00:04:31.689
DepthEye for Microsoft's original Kinect gaming
00:04:31.689 --> 00:04:35.050
system. Right. And at the time, the entire tech
00:04:35.050 --> 00:04:37.670
press got the narrative completely wrong, just
00:04:37.670 --> 00:04:39.750
like they are today with Q2II. Oh, they were
00:04:39.750 --> 00:04:42.129
so... Everyone thought Apple was going to release
00:04:42.129 --> 00:04:45.149
an Apple TV gaming console. They thought, oh,
00:04:45.230 --> 00:04:47.829
Apple bought a Kinect supplier to compete with
00:04:47.829 --> 00:04:50.660
Xbox. And then nothing. Absolute crickets for
00:04:50.660 --> 00:04:54.079
four years. From 2013 to 2017, Apple didn't say
00:04:54.079 --> 00:04:56.680
a word about PrimeSense, which, by the way, four
00:04:56.680 --> 00:04:58.800
years of silence in the tech world is an eternity.
00:04:58.980 --> 00:05:01.339
It's a lifetime. So what is happening inside
00:05:01.339 --> 00:05:03.720
Apple during a latency period like that? What
00:05:03.720 --> 00:05:05.860
are they doing with a gaming sensor? Well, you're
00:05:05.860 --> 00:05:07.779
taking a piece of hardware that was basically
00:05:07.779 --> 00:05:11.120
the size of a brick, the Kinect sensor, and you
00:05:11.120 --> 00:05:14.399
are brutalizing the supply chain to miniaturize
00:05:14.399 --> 00:05:16.850
it. Shrinking it down. Exactly. You're working
00:05:16.850 --> 00:05:19.949
out yield rates, thermal dynamics, and figuring
00:05:19.949 --> 00:05:23.649
out how to mass produce this exotic optical sensor
00:05:23.649 --> 00:05:26.250
so you can cram it into the top millimeter of
00:05:26.250 --> 00:05:28.649
a smartphone display for hundreds of millions
00:05:28.649 --> 00:05:30.910
of people. Right. And when it finally emerged
00:05:30.910 --> 00:05:34.009
from that black box in 2017, it wasn't a gaming
00:05:34.009 --> 00:05:37.470
peripheral at all. That exact same optical technology
00:05:37.470 --> 00:05:40.610
using near -infrared light to map a spaceship
00:05:40.610 --> 00:05:43.290
as the TrueDepth array. Face ID. We know it as
00:05:43.290 --> 00:05:46.259
Face ID today, yeah. Which is why... Brian Rommel's
00:05:46.259 --> 00:05:48.939
research from back then is so critical to look
00:05:48.939 --> 00:05:52.620
at today. Years before Face ID launched, he argued
00:05:52.620 --> 00:05:55.259
on Quora that PrimeSense was never a gaming leftover.
00:05:55.639 --> 00:05:57.959
He called it. He totally called it. He saw it
00:05:57.959 --> 00:06:00.160
as the first tile in a sensing stack that Apple
00:06:00.160 --> 00:06:02.980
was assembling for identity, emotion, and passwordless
00:06:02.980 --> 00:06:05.819
devices. He understood that Apple didn't buy
00:06:05.819 --> 00:06:07.660
a gaming product. Right, they bought the people.
00:06:07.939 --> 00:06:10.019
Exactly. They bought a team that had mastered
00:06:10.019 --> 00:06:12.540
a specific type of physics. And that brings us
00:06:12.540 --> 00:06:15.560
to the 2026 parallel, because the president and
00:06:15.560 --> 00:06:18.100
co -founder of a PrimeSense was Avi Edmazels.
00:06:18.279 --> 00:06:21.819
And in January 2026, this exact same guy, along
00:06:21.819 --> 00:06:24.540
with his co -founders Yonatan Wexler and Avi
00:06:24.540 --> 00:06:27.899
Barlia, sells a second company to Apple. QII.
00:06:28.180 --> 00:06:30.399
QII. They are going back inside Apple. Now, I
00:06:30.399 --> 00:06:32.660
have to stop here. Because Apple is one of the
00:06:32.660 --> 00:06:36.199
most resource -rich companies on Earth. Why drop
00:06:36.199 --> 00:06:39.740
$2 billion on the exact same founder? It's a
00:06:39.740 --> 00:06:42.459
great question. Like, is it nostalgia? Or do
00:06:42.459 --> 00:06:44.879
Maisel's and his team possess some totally unique
00:06:44.879 --> 00:06:47.079
approach to hardware that Apple's own tenured
00:06:47.079 --> 00:06:49.240
engineers just can't crack? Well, Apple could
00:06:49.240 --> 00:06:51.079
probably figure it out eventually, you know.
00:06:51.160 --> 00:06:53.939
But time is the one thing Apple can't manufacture.
00:06:54.220 --> 00:06:56.379
That makes sense. Maisel's team has a highly
00:06:56.379 --> 00:06:58.879
specialized, proven mastery over the physics
00:06:58.879 --> 00:07:01.889
of light interacting with human biology. What's
00:07:01.889 --> 00:07:04.029
fascinating here is, if you look at the lineage
00:07:04.029 --> 00:07:06.029
CrimeSense taught machines how to see the world
00:07:06.029 --> 00:07:08.750
as volume. As 3D space. Right. They used physics
00:07:08.750 --> 00:07:10.970
to solve a massive input problem, which was typing
00:07:10.970 --> 00:07:13.750
in a passcode or using a fingerprint. Face ID
00:07:13.750 --> 00:07:15.910
meant the machine finally understood your geometry.
00:07:16.230 --> 00:07:18.329
It stopped asking you to perform the physical
00:07:18.329 --> 00:07:20.889
action of pressing your thumb on a glass circle.
00:07:21.230 --> 00:07:23.470
Precisely. Now apply that exact same philosophy
00:07:23.470 --> 00:07:27.420
to Q .ai. This second company takes that mastery
00:07:27.420 --> 00:07:29.959
of optical physics and applies it to a completely
00:07:29.959 --> 00:07:32.259
different biological process. Okay, I'm tracking.
00:07:32.500 --> 00:07:34.839
Q .ai teaches machines to see the preparation
00:07:34.839 --> 00:07:37.879
of language. Just as PrimeSense eliminated the
00:07:37.879 --> 00:07:40.740
fingerprint, Q .ai provides the physics required
00:07:40.740 --> 00:07:44.120
to eliminate the wake word. Okay, so both companies
00:07:44.120 --> 00:07:47.319
solve computing input problems using light. FaceID
00:07:47.319 --> 00:07:49.620
gets rid of the fingerprint. Q .ai gets rid of
00:07:49.620 --> 00:07:51.899
HeySiri. But I want to really dig into the how
00:07:51.899 --> 00:07:54.649
here. We need to. The mechanics are crazy. Because
00:07:54.649 --> 00:07:56.410
the source material goes deep into the physics,
00:07:56.569 --> 00:07:59.350
and we need to contrast how Face ID works versus
00:07:59.350 --> 00:08:01.889
how Q .AI works. Let's start with Face ID, because
00:08:01.889 --> 00:08:03.310
most people have it in their pocket right now.
00:08:03.550 --> 00:08:06.149
Face ID uses something called structured light
00:08:06.149 --> 00:08:08.569
coding. Okay. When you look at your phone, a
00:08:08.569 --> 00:08:11.930
dot projector fires about 30 ,000 tiny dots of
00:08:11.930 --> 00:08:13.990
near -infrared light onto your face. And because
00:08:13.990 --> 00:08:16.750
my face isn't flat, I've got a nose, cheekbones,
00:08:16.829 --> 00:08:19.509
eye sockets. That 3D geometry warps the pattern
00:08:19.509 --> 00:08:22.649
of those 30 ,000 dots, right? Precisely. Your
00:08:22.649 --> 00:08:26.529
anatomy distorts the grid. Then a standard CMOS
00:08:26.529 --> 00:08:29.149
sensor, which is essentially just the basic digital
00:08:29.149 --> 00:08:31.870
camera chip that's in every smartphone, reads
00:08:31.870 --> 00:08:34.289
that distortion. So it's basically taking a picture
00:08:34.289 --> 00:08:37.289
of the bent dots. Yeah, and clever software calculates
00:08:37.289 --> 00:08:40.090
the depth based on how the dot's bent. But here's
00:08:40.090 --> 00:08:43.850
the key. Face ID maps your face as a rigid, static
00:08:43.850 --> 00:08:47.450
3D object. It's looking at the macroscopic structure
00:08:47.450 --> 00:08:50.159
of your head. I'm trying to visualize this. Let's
00:08:50.159 --> 00:08:53.139
say Face ID is like a satellite taking a high
00:08:53.139 --> 00:08:55.399
-resolution photograph of a frozen lake. Oh,
00:08:55.460 --> 00:08:57.580
that's a good analogy. It maps the surface perfectly.
00:08:57.759 --> 00:09:00.500
It sees the static shape of the ice, the ridges,
00:09:00.519 --> 00:09:03.139
the shoreline. It creates a perfect model of
00:09:03.139 --> 00:09:05.580
a large, rigid object. That's a great baseline.
00:09:05.700 --> 00:09:08.120
Now, Q .AI is not looking at the static ice.
00:09:08.299 --> 00:09:11.120
Right. Q .AI is the next resolution down. Much,
00:09:11.120 --> 00:09:14.360
much further down. Q .AI doesn't project 30 ,000
00:09:14.360 --> 00:09:16.840
dots over your whole face. It shines a very targeted
00:09:16.840 --> 00:09:19.480
beam of coherence. light onto a very small patch
00:09:19.480 --> 00:09:21.860
of your facial skin. Wait, pause. What makes
00:09:21.860 --> 00:09:24.320
light coherent? Why does that matter? Good question.
00:09:24.639 --> 00:09:27.259
Think of the light bulb in your lamp. It throws
00:09:27.259 --> 00:09:29.779
light out in every direction, totally disorganized.
00:09:29.879 --> 00:09:32.299
That's incoherent. Like just splashing water
00:09:32.299 --> 00:09:35.850
everywhere. Exactly. Coherent light, like a laser,
00:09:35.990 --> 00:09:38.169
consists of light waves that are marching in
00:09:38.169 --> 00:09:40.509
perfect lockstep with one another. They are highly
00:09:40.509 --> 00:09:42.909
organized. Okay, got it. When you shine coherent
00:09:42.909 --> 00:09:45.750
light onto a surface like human skin and it bounces
00:09:45.750 --> 00:09:48.429
back to a detector, it creates a very specific
00:09:48.429 --> 00:09:51.789
interference pattern called a speckle. A speckle.
00:09:51.889 --> 00:09:54.669
Okay, so the coherent light bounces off my skin
00:09:54.669 --> 00:09:56.929
and creates this speckle pattern on the sensor.
00:09:57.330 --> 00:09:59.850
Where does the speech part come in? This is where
00:09:59.850 --> 00:10:02.409
we cross from physics into biology. Buckle up.
00:10:02.610 --> 00:10:04.950
The source analysis explains that when you are
00:10:04.950 --> 00:10:07.409
about to speak before air ever leaves your lungs,
00:10:07.610 --> 00:10:10.169
the muscles that form a vowel or a consonant
00:10:10.169 --> 00:10:13.230
begin to twitch. But these are tiny. Right? Oh,
00:10:13.409 --> 00:10:15.269
we're talking about twitches on the scale of
00:10:15.269 --> 00:10:18.350
tens of microns. A micron is one thousandth of
00:10:18.350 --> 00:10:20.909
a millimeter. It is completely invisible to the
00:10:20.909 --> 00:10:23.090
human eye. But not invisible to the coherent
00:10:23.090 --> 00:10:26.470
light. Exactly. When that microscopic muscle
00:10:26.470 --> 00:10:29.090
twitch happens under your skin, it changes the
00:10:29.090 --> 00:10:31.549
surface of your skin by a fraction of hair's
00:10:31.549 --> 00:10:34.710
width. But because the light is coherent, that
00:10:34.710 --> 00:10:37.549
tiny shift violently disrupts the speckle pattern
00:10:37.549 --> 00:10:39.950
bouncing back to the sensor. Whoa! The sensor
00:10:39.950 --> 00:10:43.009
reads that shifting speckle. Then... A highly
00:10:43.009 --> 00:10:46.190
trained neural network maps that specific speckle
00:10:46.190 --> 00:10:49.350
shift onto phonemes. And phonemes are just the
00:10:49.350 --> 00:10:51.409
distinct sounds that make up a language, like
00:10:51.409 --> 00:10:54.110
the K sound in cat or the B sound in bat. Right.
00:10:54.250 --> 00:10:56.590
The neural network pieces those phones together
00:10:56.590 --> 00:10:59.389
into words and ultimately into your intention
00:10:59.389 --> 00:11:02.129
to speak before the speech has ever become an
00:11:02.129 --> 00:11:04.669
audible sound. That is mind -bending. It is reading
00:11:04.669 --> 00:11:07.330
the biological motor plan of language. And it
00:11:07.330 --> 00:11:09.570
does this entirely optically, from a distance.
00:11:10.029 --> 00:11:12.929
No electrodes taped to your throat. No physical
00:11:12.929 --> 00:11:15.509
contact with your skin. So going back to my frozen
00:11:15.509 --> 00:11:18.570
lake analogy, if Face ID is the satellite mapping
00:11:18.570 --> 00:11:20.590
the frozen static surface of the ice, then Q
00:11:20.590 --> 00:11:23.149
.AI is like shining a laser onto the water before
00:11:23.149 --> 00:11:25.649
it freezes, watching the microscopic ripples
00:11:25.649 --> 00:11:27.909
and currents beneath the surface to predict exactly
00:11:27.909 --> 00:11:30.710
where and how a wave is going to break. It's
00:11:30.710 --> 00:11:32.990
reading the energy before the event happens.
00:11:33.210 --> 00:11:35.629
That captures the physics perfectly. Yeah. And
00:11:35.629 --> 00:11:38.409
this physical leap is staggering. It moves the
00:11:38.409 --> 00:11:41.730
entire computing industry from reading static
00:11:41.730 --> 00:11:45.279
geometry anatomy to... reading living dynamic
00:11:45.279 --> 00:11:48.700
biological processes in real time. We are shifting
00:11:48.700 --> 00:11:51.980
to physiology. Exactly. And this isn't just some
00:11:51.980 --> 00:11:54.600
theoretical white paper either. According to
00:11:54.600 --> 00:11:57.419
Rommel's analysis, Apple has already mapped this
00:11:57.419 --> 00:12:00.399
out meticulously in a vast web of patents that
00:12:00.399 --> 00:12:03.340
they now own. They have the blueprint. And that
00:12:03.340 --> 00:12:06.039
blueprint is how you justify a $2 billion price
00:12:06.039 --> 00:12:10.059
tag. The analysis points out that Q .AI wasn't
00:12:10.059 --> 00:12:12.940
just a single parlor trick. It is an entire product
00:12:12.940 --> 00:12:15.700
architecture written in patent claims. Right.
00:12:15.799 --> 00:12:17.440
It's not just a sensor. It's a whole system.
00:12:17.580 --> 00:12:19.379
When you strip away the patent numbers and just
00:12:19.379 --> 00:12:22.179
look at what these documents describe, a massive,
00:12:22.320 --> 00:12:25.320
highly disruptive product roadmap emerges. Let's
00:12:25.320 --> 00:12:26.779
walk through the architecture because I don't
00:12:26.779 --> 00:12:28.539
want to just list off a bunch of eight -digit
00:12:28.539 --> 00:12:30.220
patent registry numbers that nobody's going to
00:12:30.220 --> 00:12:32.120
remember. Let's look at what these patents actually
00:12:32.120 --> 00:12:34.120
do. do. Yeah, the practical application. The
00:12:34.120 --> 00:12:36.320
first one we have to talk about is what I'll
00:12:36.320 --> 00:12:38.840
call the skin transcription patent. This is the
00:12:38.840 --> 00:12:42.450
foundational one. The claim describes deciphering
00:12:42.450 --> 00:12:45.250
silent speech by shining coherent light onto
00:12:45.250 --> 00:12:48.029
the skin, reading the speckle coming in, and
00:12:48.029 --> 00:12:50.769
generating transcribed language out. It's basically
00:12:50.769 --> 00:12:52.950
the core mechanic we just talked about. Exactly.
00:12:52.950 --> 00:12:56.029
It explicitly covers people articulating words
00:12:56.029 --> 00:12:58.370
without making any physical contact with the
00:12:58.370 --> 00:13:00.710
sensor. Okay, so that's the core physics. But
00:13:00.710 --> 00:13:02.950
then the next group of patents answers the obvious
00:13:02.950 --> 00:13:05.190
question, where does this laser actually live
00:13:05.190 --> 00:13:08.830
on my body? And that brings us to the earbud
00:13:08.830 --> 00:13:11.169
housing patent. This is where it gets real for
00:13:11.169 --> 00:13:13.740
the consumer. The patents detail determining
00:13:13.740 --> 00:13:16.799
speech from skin movements using a housing supported
00:13:16.799 --> 00:13:20.360
by the human ear. So AirPods. This is the AirPods
00:13:20.360 --> 00:13:23.320
claim set. It tells us exactly what hardware
00:13:23.320 --> 00:13:25.860
ecosystem this sensor is destined for. Which
00:13:25.860 --> 00:13:28.279
makes perfect sense physically. The stem of an
00:13:28.279 --> 00:13:30.940
AirPod points right down at your cheek and jaw
00:13:30.940 --> 00:13:33.460
hinge. It's in the perfect position to shine
00:13:33.460 --> 00:13:35.440
that coherent light. It couldn't be positioned
00:13:35.440 --> 00:13:37.840
better. But here is the patent that absolutely
00:13:37.840 --> 00:13:40.679
blew my mind. Let's call it the intent patent.
00:13:41.210 --> 00:13:44.590
Oh, this one is wild. It's literally titled interpreting
00:13:44.590 --> 00:13:48.190
words prior to vocalization alongside another
00:13:48.190 --> 00:13:52.309
for facial movements, waking up a wearable. Interpreting
00:13:52.309 --> 00:13:55.409
words prior to vocalization. That is the death
00:13:55.409 --> 00:13:58.009
of the wake word right there. The early muscular
00:13:58.009 --> 00:14:00.110
recruitment of your face getting ready to speak
00:14:00.110 --> 00:14:03.490
is the new. Haste, Siri. Think about how profound
00:14:03.490 --> 00:14:06.049
that is. You don't announce your intention to
00:14:06.049 --> 00:14:08.529
a machine anymore. The machine observes your
00:14:08.529 --> 00:14:10.889
biology, preparing to act. And they pair that
00:14:10.889 --> 00:14:13.450
with what we should call the privacy output patent.
00:14:13.990 --> 00:14:16.250
Because if a machine can read your silent query,
00:14:16.549 --> 00:14:18.769
it needs to give you a silent answer. Right.
00:14:18.830 --> 00:14:20.470
You don't want it shouting the answer out loud.
00:14:20.840 --> 00:14:23.059
This patent covers providing private answers
00:14:23.059 --> 00:14:26.460
to non -vocal questions, routing the audio discreetly
00:14:26.460 --> 00:14:28.620
back into your earbud rather than blasting it
00:14:28.620 --> 00:14:30.659
out of a phone speaker. Okay, I have to jump
00:14:30.659 --> 00:14:33.139
in here with a very practical, real -world concern
00:14:33.139 --> 00:14:35.559
based on the source text. Sure, go ahead. If
00:14:35.559 --> 00:14:38.100
I have this future AirPod in my ear and it is
00:14:38.100 --> 00:14:40.580
constantly reading the microscopic muscle twitches
00:14:40.580 --> 00:14:42.799
of my cheek and jaw, what happens if I'm just
00:14:42.799 --> 00:14:45.279
chewing a piece of gum? Or eating a sandwich?
00:14:45.500 --> 00:14:47.759
Or what if I smile at someone on the street or
00:14:47.759 --> 00:14:49.679
just have a random nervous twitch? This is the
00:14:49.679 --> 00:14:52.000
first thing everyone asks. Is my phone going
00:14:52.000 --> 00:14:54.720
to constantly misinterpret that as a word and
00:14:54.720 --> 00:14:57.879
fire off gibberish texts to my boss? That is
00:14:57.879 --> 00:15:00.820
the exact hurdle that separates a neat lab experiment
00:15:00.820 --> 00:15:04.320
from a shipping consumer product. But the patent
00:15:04.320 --> 00:15:06.740
portfolio accounts for this. They have a whole
00:15:06.740 --> 00:15:09.500
set of claims covering threshold intensity. What
00:15:09.500 --> 00:15:11.769
does that mean in plain English? Threshold intensity.
00:15:12.169 --> 00:15:13.970
It means the neural network isn't just looking
00:15:13.970 --> 00:15:17.090
for any movement. Chewing gum creates a massive
00:15:17.090 --> 00:15:20.750
rhythmic macro movement of the jaw. Smiling stretches
00:15:20.750 --> 00:15:23.990
the skin in a very specific global way. So the
00:15:23.990 --> 00:15:25.830
sensor can tell the difference. The system is
00:15:25.830 --> 00:15:27.830
trained to filter those out as optical noise.
00:15:28.490 --> 00:15:30.830
It is looking for the highly specific, incredibly
00:15:30.830 --> 00:15:33.710
subtle micro sequencing of a motor plan forming
00:15:33.710 --> 00:15:36.610
a phone. It knows the difference between a chew
00:15:36.610 --> 00:15:39.090
and a consonant. So when you synthesize all of
00:15:39.090 --> 00:15:42.070
these pieces. The skin transcription, the earbud
00:15:42.070 --> 00:15:45.269
housing, the pre -vocalization intent, the privacy
00:15:45.269 --> 00:15:48.029
output, the threshold filters. You realize Apple
00:15:48.029 --> 00:15:50.370
didn't buy a microphone company. Not at all.
00:15:50.470 --> 00:15:53.570
They bought an entirely new input layer. It's
00:15:53.570 --> 00:15:55.889
a computer interface that no longer requires
00:15:55.889 --> 00:15:58.970
the physical act of spoken speech. It bypasses
00:15:58.970 --> 00:16:02.009
the air entirely. But having the patents in a
00:16:02.009 --> 00:16:04.509
vault in Cupertino is one thing. How does this
00:16:04.509 --> 00:16:07.429
actually integrate into the Apple devices that
00:16:07.429 --> 00:16:09.649
you, the listener, are carrying in your pocket?
00:16:10.299 --> 00:16:16.320
That's the billion -dollar question. I want to
00:16:16.320 --> 00:16:18.340
pause on the word connectionless because we aren't
00:16:18.340 --> 00:16:20.399
talking about losing your Wi -Fi signal here.
00:16:20.539 --> 00:16:22.779
No, connectionless here refers to the user experience.
00:16:23.320 --> 00:16:25.820
Think about how every voice assistant has functioned
00:16:25.820 --> 00:16:28.899
since Siri launched in 2011. You have to establish
00:16:28.899 --> 00:16:30.639
a connection. You have to initiate it. Right.
00:16:30.740 --> 00:16:33.580
You shout a wake word across a room, or you unlock
00:16:33.580 --> 00:16:35.899
your phone, open an app, and hold down a button.
00:16:36.570 --> 00:16:39.389
You perform an action to open an acoustic session.
00:16:39.590 --> 00:16:42.350
But with this tech, I don't have to pair anything
00:16:42.350 --> 00:16:45.070
new. I don't hold a button. I don't even have
00:16:45.070 --> 00:16:47.389
to finish the word I'm thinking of. The device
00:16:47.389 --> 00:16:50.490
is already optically coupled to my face. To truly
00:16:50.490 --> 00:16:53.429
understand how revolutionary this is, the source
00:16:53.429 --> 00:16:55.610
material breaks down something called the speech
00:16:55.610 --> 00:16:58.799
cascade. I think this is crucial because most
00:16:58.799 --> 00:17:01.259
people misunderstand what speech actually is.
00:17:01.379 --> 00:17:03.960
Right. This totally flipped my perspective. We
00:17:03.960 --> 00:17:07.440
think of speech as a sound event. But sound is
00:17:07.440 --> 00:17:10.460
literally just the final exhaust of a much longer
00:17:10.460 --> 00:17:13.480
process. The very end of the line. The text outlines
00:17:13.480 --> 00:17:16.539
the cascade like this. First, there's the intent,
00:17:16.839 --> 00:17:19.380
the thought in your brain. Second, your brain
00:17:19.380 --> 00:17:22.460
creates a motor planning signal. Third, there
00:17:22.460 --> 00:17:24.759
is the recruitment of your facial and laryngeal
00:17:24.759 --> 00:17:27.579
muscles to execute that plan. And finally, fourth,
00:17:27.779 --> 00:17:30.039
air gets pushed from your lungs through those
00:17:30.039 --> 00:17:32.339
muscles to create sound. Old voice technology
00:17:32.339 --> 00:17:34.920
only interacts with step four, the air. Here's
00:17:34.920 --> 00:17:36.700
where it gets really interesting. Let me use
00:17:36.700 --> 00:17:38.740
another analogy. Think about a baseball game.
00:17:38.880 --> 00:17:41.500
The voice tech we use today only reacts once
00:17:41.500 --> 00:17:43.539
the ball has left the pitcher's hand and is flying
00:17:43.539 --> 00:17:46.029
through the air toward home plate. The microphone
00:17:46.029 --> 00:17:48.089
has to wait to catch the sound wave in flight.
00:17:48.250 --> 00:17:51.809
Sure. But QAI is reading the pitcher's windup.
00:17:51.950 --> 00:17:54.089
It reads the muscles in the pitcher's shoulder
00:17:54.089 --> 00:17:56.789
before the ball is ever released. It knows exactly
00:17:56.789 --> 00:17:59.049
what pitch is coming, how fast it'll be, and
00:17:59.049 --> 00:18:02.609
where it's aimed. That's spot on. Q .AI positions
00:18:02.609 --> 00:18:06.690
its sensors at step three of the cascade. It
00:18:06.690 --> 00:18:09.549
is, by definition, a brain -computer interface.
00:18:10.200 --> 00:18:14.059
A BCI. Wait, a BCI? Like Neuralink? But nobody
00:18:14.059 --> 00:18:16.200
is drilling holes in my head for this. Exactly.
00:18:16.400 --> 00:18:20.059
It's a non -surgical, non -invasive BCI. It reads
00:18:20.059 --> 00:18:22.299
the biological exhaust of your brain's motor
00:18:22.299 --> 00:18:24.480
plan without ever breaking the skin or forcing
00:18:24.480 --> 00:18:26.720
you to wear a skull cap of electrodes. That is
00:18:26.720 --> 00:18:29.660
wild. And when you map this onto Apple's current
00:18:29.660 --> 00:18:31.940
hardware stack, the elegance of the integration
00:18:31.940 --> 00:18:34.700
is undeniable. If we connect this to the bigger
00:18:34.700 --> 00:18:37.150
picture, let's start with the AirPods. We talked
00:18:37.150 --> 00:18:39.269
about the earbud patent earlier. The stem of
00:18:39.269 --> 00:18:41.150
the AirPod points right at the jaw. But the text
00:18:41.150 --> 00:18:43.670
mentions the earbud acting as a local interferometer.
00:18:43.809 --> 00:18:45.329
You're going to have to explain that one. An
00:18:45.329 --> 00:18:47.329
interferometer is just an instrument that bounces
00:18:47.329 --> 00:18:49.769
light waves off something and measures the interference
00:18:49.769 --> 00:18:51.809
pattern when the light comes back. Okay, so it
00:18:51.809 --> 00:18:53.990
measures interference. It's a way to measure
00:18:53.990 --> 00:18:56.750
incredibly tiny changes in distance or surface.
00:18:57.410 --> 00:19:00.670
So imagine the next generation of AirPods has
00:19:00.670 --> 00:19:03.549
a microscopic coherent light emitter and detector
00:19:03.549 --> 00:19:07.250
built into the plastic of the stem. Your AirPod
00:19:07.250 --> 00:19:09.630
is now an optical interferometer, constantly
00:19:09.630 --> 00:19:11.710
measuring the speckle on your cheek. Okay, so
00:19:11.710 --> 00:19:14.829
the AirPod is the primary sensor. But Apple's
00:19:14.829 --> 00:19:17.210
whole philosophy is ecosystem redundancy, right?
00:19:17.349 --> 00:19:19.650
The devices talk to each other. Right, which
00:19:19.650 --> 00:19:22.190
brings in the iPhone's TrueDepth camera, the
00:19:22.190 --> 00:19:24.990
Face ID sensor we talked about earlier. Oh, circling
00:19:24.990 --> 00:19:27.809
back to PrimeSense. Exactly. PrimeSense gave
00:19:27.809 --> 00:19:30.509
Apple the ability to throw structured light to
00:19:30.509 --> 00:19:33.329
verify your identity. QDAT AI gives them the
00:19:33.329 --> 00:19:35.349
knowledge to throw coherent light to read your
00:19:35.349 --> 00:19:37.509
language. Imagine you are holding your phone
00:19:37.509 --> 00:19:39.589
up, looking at the screen. Like scrolling or
00:19:39.589 --> 00:19:42.190
whatever. Right. The camera in the dynamic island
00:19:42.190 --> 00:19:45.009
can act as a secondary sensor, reading the speckle
00:19:45.009 --> 00:19:46.869
on your lips and face while the AirPod reads
00:19:46.869 --> 00:19:50.410
your cheek. Two optical opinions beat one. It
00:19:50.410 --> 00:19:53.569
helps disambiguate silent phonemes, but... All
00:19:53.569 --> 00:19:56.130
of this biological data, these constant speckle
00:19:56.130 --> 00:19:58.970
shifts, have to be processed instantly. Milliseconds.
00:19:58.990 --> 00:20:01.609
And this is where Apple's obsession with proprietary
00:20:01.609 --> 00:20:05.089
silicon comes in. The neural engine. This is
00:20:05.089 --> 00:20:07.490
the linchpin of the whole operation. Yeah. The
00:20:07.490 --> 00:20:09.789
patent descriptions specifically note that these
00:20:09.789 --> 00:20:12.650
silent speech models are highly optimized and
00:20:12.650 --> 00:20:14.970
small enough to run locally on the device's neural
00:20:14.970 --> 00:20:17.140
processing unit. Why does running it locally
00:20:17.140 --> 00:20:19.339
matter so much? Why not just send the speckled
00:20:19.339 --> 00:20:21.819
data to the cloud like ChatGPT does with text?
00:20:22.079 --> 00:20:24.900
Because of latency and privacy. If you have to
00:20:24.900 --> 00:20:27.380
beam microscopic muscle data to a server in Virginia
00:20:27.380 --> 00:20:29.640
to figure out what word you're mouthing, the
00:20:29.640 --> 00:20:31.740
delay breaks the magic. You'd be waiting for
00:20:31.740 --> 00:20:34.440
your own thought to load. Exactly. And more importantly,
00:20:34.599 --> 00:20:37.680
this perfectly aligns with Apple's 2026 on -device
00:20:37.680 --> 00:20:40.900
-first AI posture. You simply cannot sell a consumer
00:20:40.900 --> 00:20:43.700
a device that constantly reads their pre -vocalized
00:20:43.700 --> 00:20:46.220
biological intent if that data leaves the hardware.
00:20:46.420 --> 00:20:49.259
It has to be an absolute closed loop. And we
00:20:49.259 --> 00:20:51.980
can't forget about Vision Pro or whatever smart
00:20:51.980 --> 00:20:55.250
glasses Apple inevitably releases. A headset
00:20:55.250 --> 00:20:57.509
is inherently locked onto your face. Oh, the
00:20:57.509 --> 00:20:59.990
wearables aspect is huge. The source notes that
00:20:59.990 --> 00:21:02.430
Apple is already predicting user clicks based
00:21:02.430 --> 00:21:05.529
on pupil movement. Adding a cheek reading channel
00:21:05.529 --> 00:21:07.650
to a pair of smart glasses is just geometry.
00:21:08.170 --> 00:21:11.130
Pupil prediction was the dress rehearsal reading
00:21:11.130 --> 00:21:14.150
pre -speech intent is the main event. And beyond
00:21:14.150 --> 00:21:16.470
just convenience, think about the security implications.
00:21:16.890 --> 00:21:20.190
This entire stack creates a continuous authentication
00:21:20.190 --> 00:21:23.579
layer. How so? Well, Face ID occasionally asks,
00:21:23.700 --> 00:21:26.700
is this the correct rigid sculpture of a face?
00:21:27.180 --> 00:21:31.000
But a QII -enabled device constantly asks, is
00:21:31.000 --> 00:21:33.920
this the correct living dynamic muscle signature
00:21:33.920 --> 00:21:36.660
at this exact millisecond? It's the ultimate
00:21:36.660 --> 00:21:39.160
proof of liveness. Spoofing an optical speckle
00:21:39.160 --> 00:21:41.599
of a human muscle twitching in real time is exponentially
00:21:41.599 --> 00:21:44.319
harder than holding up a 3D -printed mask. Exactly.
00:21:44.730 --> 00:21:46.650
But let's pivot to the human element of this
00:21:46.650 --> 00:21:48.230
because this incredible hardware integration
00:21:48.230 --> 00:21:51.130
solves a massive behavioral problem that has
00:21:51.130 --> 00:21:53.509
played voice tech for over a decade. Right. The
00:21:53.509 --> 00:21:55.130
source analysis calls it the end of theater.
00:21:55.390 --> 00:21:58.289
Yes, because let's be honest with ourselves.
00:21:58.549 --> 00:22:01.109
The reason voice assistants like Siri, Alexa,
00:22:01.349 --> 00:22:03.869
or Google haven't become our primary computing
00:22:03.869 --> 00:22:06.710
interface isn't just because they sometimes misunderstand
00:22:06.710 --> 00:22:09.690
us. No, the accuracy is actually pretty good
00:22:09.690 --> 00:22:12.150
now. The real failure was the social performance
00:22:12.150 --> 00:22:14.970
required to use them. The social friction is
00:22:14.970 --> 00:22:18.009
immense. People simply will not sit in a quiet
00:22:18.009 --> 00:22:20.569
open plan office and say, hey, Siri, summarize
00:22:20.569 --> 00:22:22.730
the quarterly financials. Or they won't stand
00:22:22.730 --> 00:22:25.130
on a crowded train and say, hey, Siri, text my
00:22:25.130 --> 00:22:27.529
doctor about my prescription. Right. The theater
00:22:27.529 --> 00:22:31.130
of voice computing is inherently awkward. Deeply
00:22:31.130 --> 00:22:33.569
unprivate. But with connectionless intent reading,
00:22:33.869 --> 00:22:36.630
users will transition to simply mouthing their
00:22:36.630 --> 00:22:39.450
queries. You don't whisper. You don't push air.
00:22:39.730 --> 00:22:41.549
You just move the muscles. You just let your
00:22:41.549 --> 00:22:43.970
facial muscles form the sentence silently and
00:22:43.970 --> 00:22:46.109
the earbud catches the motor plan and confirms
00:22:46.109 --> 00:22:49.500
the action privately in your ear. Remillel argues
00:22:49.500 --> 00:22:51.539
in the text that the requirement for our thoughts
00:22:51.539 --> 00:22:53.740
to become physical air pressure before a computer
00:22:53.740 --> 00:22:56.259
could understand them was always a bug in computing.
00:22:56.559 --> 00:22:58.380
We just accepted it because we didn't have the
00:22:58.380 --> 00:23:01.099
physics to bypass it yet. It's a profound observation.
00:23:01.579 --> 00:23:04.200
However, as analysts, we have a responsibility
00:23:04.200 --> 00:23:07.359
to look at the real -world caveats here. The
00:23:07.359 --> 00:23:10.059
distance between a pristine patent filing and
00:23:10.059 --> 00:23:12.980
a shipping consumer product is a minefield. Oh,
00:23:13.019 --> 00:23:15.490
it's a massive gap. We have to do a reality check.
00:23:15.670 --> 00:23:17.990
Right, let's do a reality check. The analysis
00:23:17.990 --> 00:23:20.609
outlines some very real constraints. First off,
00:23:20.730 --> 00:23:23.750
historically, silent speech interfaces built
00:23:23.750 --> 00:23:26.509
in university labs have totally collapsed outside
00:23:26.509 --> 00:23:28.690
of speaker -specific training. They're incredibly
00:23:28.690 --> 00:23:31.750
fragile. Like, they work flawlessly for the one
00:23:31.750 --> 00:23:34.589
grad student who spent 100 hours calibrating
00:23:34.589 --> 00:23:37.549
the sensor to their specific jaw structure, but
00:23:37.549 --> 00:23:39.869
they fail the moment someone else puts it on.
00:23:39.990 --> 00:23:42.619
And that's in a lab. Reading an optical speckle
00:23:42.619 --> 00:23:44.880
on a cheek in a perfectly lit, climate -controlled
00:23:44.880 --> 00:23:47.319
environment is one thing. Trying to read that
00:23:47.319 --> 00:23:49.559
same speckle on a sweaty runner's cheek while
00:23:49.559 --> 00:23:51.599
they are jogging down a windy street corner in
00:23:51.599 --> 00:23:53.839
Chicago is an entirely different physics problem.
00:23:54.039 --> 00:23:56.980
The environmental noise is massive. Wind, sweat,
00:23:57.240 --> 00:24:00.299
movement. Plus, the financial analysts tracking
00:24:00.299 --> 00:24:02.740
this deal still disagree on the final acquisition
00:24:02.740 --> 00:24:06.349
price by up to half a billion dollars. Which
00:24:06.349 --> 00:24:08.789
just shows how deeply under wraps this technology
00:24:08.789 --> 00:24:12.130
still is. Apple hasn't officially confirmed exactly
00:24:12.130 --> 00:24:15.650
how or when this will ship. Which brings us back
00:24:15.650 --> 00:24:18.069
to the PrimeSense timeline. Remember, that took
00:24:18.069 --> 00:24:20.390
four years of silence before Face ID shipped.
00:24:20.549 --> 00:24:23.130
Right. The engineering required to miniaturize
00:24:23.130 --> 00:24:26.109
a coherent light interferometer into an AirPod
00:24:26.109 --> 00:24:29.170
stem while maintaining battery life is monumental.
00:24:29.450 --> 00:24:32.769
If the deal closed in 2026, we shouldn't be surprised
00:24:32.769 --> 00:24:34.990
if we don't see this fully integrated until 2030.
00:24:35.529 --> 00:24:38.630
So as we look toward that 2030 horizon, what
00:24:38.630 --> 00:24:40.450
does this actually mean for us? I have to be
00:24:40.450 --> 00:24:42.829
honest, it sets off some massive privacy alarms
00:24:42.829 --> 00:24:45.630
for me. Understandably so. A sensor sitting on
00:24:45.630 --> 00:24:48.210
my face, constantly reading my pre -speech. Is
00:24:48.210 --> 00:24:50.170
this the ultimate helpful assistant that knows
00:24:50.170 --> 00:24:52.549
what I need with zero friction? Or is this the
00:24:52.549 --> 00:24:54.670
ultimate surveillance device if that invisible
00:24:54.670 --> 00:24:57.210
laser stays on when I don't want it to? It is
00:24:57.210 --> 00:25:00.210
entirely valid to feel uneasy about it. A sensor
00:25:00.210 --> 00:25:02.490
capable of reading intent before vocalization
00:25:02.490 --> 00:25:05.089
is one of the most potent technologies ever conceived.
00:25:05.329 --> 00:25:07.630
Right. Whether it is a tool or a surveillance
00:25:07.630 --> 00:25:09.829
monitor depends entirely on the architecture.
00:25:10.329 --> 00:25:13.349
Does Apple hold the model locally? Does the data
00:25:13.349 --> 00:25:15.849
stay on the neural engine? Can the user physically
00:25:15.849 --> 00:25:18.170
sever the optical connection? This is why that
00:25:18.170 --> 00:25:20.150
local processing we talked about earlier is the
00:25:20.150 --> 00:25:22.589
absolute linchpin. If this data went to the cloud,
00:25:22.650 --> 00:25:24.569
it would be a non -starter. Society would reject
00:25:24.569 --> 00:25:27.859
it. Precisely. The local processing is the only
00:25:27.859 --> 00:25:30.059
thing that gives this technology a chance to
00:25:30.059 --> 00:25:33.799
be accepted as a secure, empowering tool. But
00:25:33.799 --> 00:25:36.960
the reality remains, Apple now owns both the
00:25:36.960 --> 00:25:39.660
camera that proved you were you and the camera
00:25:39.660 --> 00:25:41.279
that can hear the sentence, you haven't even
00:25:41.279 --> 00:25:43.680
spoken yet. Wow. They own the identity layer,
00:25:43.819 --> 00:25:46.319
and now they own the intent layer. Back in 2013,
00:25:46.599 --> 00:25:48.460
the entire industry wondered why Apple would
00:25:48.460 --> 00:25:50.660
pay hundreds of millions of dollars for a Kinect
00:25:50.660 --> 00:25:53.400
gaming supplier. Four years later, the answer
00:25:53.400 --> 00:25:56.119
was sitting on the front of every iPhone in the
00:25:56.119 --> 00:25:58.819
world. And in 2026, the question is why Apple
00:25:58.819 --> 00:26:01.279
would pay $2 billion for a company with no consumer
00:26:01.279 --> 00:26:04.059
product and a laser aimed at a cheek. And the
00:26:04.059 --> 00:26:07.180
answer is already sitting in your ear. The AirPods
00:26:07.180 --> 00:26:09.619
are the connection. The intent is already written
00:26:09.619 --> 00:26:12.059
on your face. The only thing left for Apple to
00:26:12.059 --> 00:26:14.569
remove is the requirement that you speak. That's
00:26:14.569 --> 00:26:17.809
exactly it. Which leaves me with one final kind
00:26:17.809 --> 00:26:20.269
of provocative thought for you to chew on after
00:26:20.269 --> 00:26:23.470
this deep dive is over. We spent a lot of time
00:26:23.470 --> 00:26:26.130
talking about the speech cascade today, intent,
00:26:26.569 --> 00:26:29.410
motor planning, muscle recruitment, and finally,
00:26:29.549 --> 00:26:33.230
air. But what happens to human self -correction?
00:26:33.549 --> 00:26:35.829
Oh, that's an interesting angle. I mean, we have
00:26:35.829 --> 00:26:38.190
all had those moments where our brain fully intends
00:26:38.190 --> 00:26:40.750
to say something. Maybe you're angry. Your jaw
00:26:40.750 --> 00:26:43.529
muscles literally begin to form the words. But
00:26:43.529 --> 00:26:45.930
your social filter kicks in at the very last
00:26:45.930 --> 00:26:48.589
possible millisecond, and you bite your tongue.
00:26:48.730 --> 00:26:50.650
You stop yourself before making a sound. You
00:26:50.650 --> 00:26:53.710
hold back the thunder. It's true. We rely heavily
00:26:53.710 --> 00:26:55.890
on that acoustic delay to save us from our worst
00:26:55.890 --> 00:26:58.700
impulses. Exactly. So if Apple's new technology
00:26:58.700 --> 00:27:01.460
acts on our pre -vocalization, if it reads that
00:27:01.460 --> 00:27:03.900
microscopic muscle twitch that happens before
00:27:03.900 --> 00:27:06.900
our filter catches it, will our devices execute
00:27:06.900 --> 00:27:09.980
commands or send private messages that we actually
00:27:09.980 --> 00:27:11.980
changed our minds about a fraction of a second
00:27:11.980 --> 00:27:14.200
later? It's a terrifying thought. What happens
00:27:14.200 --> 00:27:17.220
to us socially and psychologically when the gap
00:27:17.220 --> 00:27:19.960
between impulse and action is completely erased?
00:27:20.259 --> 00:27:21.920
It makes you wonder if living in the thunder
00:27:21.920 --> 00:27:24.380
wasn't so bad after all. Thank you for joining
00:27:24.380 --> 00:27:26.279
us on this deep dive into the source material.
00:27:26.500 --> 00:27:45.150
We'll catch you on the next one. Technology and
00:27:45.150 --> 00:27:49.750
history, he connects them all for you Brian Romer
00:27:49.750 --> 00:27:52.049
leads the way with research deep and true Denology
00:27:52.049 --> 00:27:54.230
and history, he connects them all for you Ancient
00:27:54.230 --> 00:27:56.730
sparks to modern code, he traces every thread
00:28:29.519 --> 00:28:34.079
But humanity moves on. Love equation for AI.
00:28:34.660 --> 00:28:39.299
Zero human companies building from the start.
00:28:40.579 --> 00:28:44.680
Thermodynamic wages earning the key. Secrets
00:28:44.680 --> 00:28:45.640
we will read.
00:00:00.000 --> 00:00:02.580
This is the Read Multiplex podcast, and this
00:00:02.580 --> 00:00:05.440
is a special analysis of a company by Apple that
00:00:05.440 --> 00:00:08.279
only Brian Rommel could give. If you like his
00:00:08.279 --> 00:00:10.320
work or get anything out of it, become a member
00:00:10.320 --> 00:00:12.619
or at least buy him a coffee. Absolutely. Yeah.
00:00:12.740 --> 00:00:16.199
So with that out of the way, I mean, what if
00:00:16.199 --> 00:00:18.940
I told you Apple just spent up to two billion
00:00:18.940 --> 00:00:21.039
dollars to make sure Siri can read your mind
00:00:21.039 --> 00:00:23.320
before you even open your mouth? It sounds totally
00:00:23.320 --> 00:00:25.359
insane when you say it like that out loud. Right.
00:00:25.539 --> 00:00:28.760
It sounds like pure sci -fi. But today we're
00:00:28.760 --> 00:00:31.280
going into some incredibly strange, fascinating
00:00:31.280 --> 00:00:35.219
territory because, well, we're so used to the
00:00:35.219 --> 00:00:37.880
idea that action in our physical world is like
00:00:37.880 --> 00:00:40.880
thunder. Yeah, the thunder analogy is perfect
00:00:40.880 --> 00:00:42.840
for this. You see a lightning strike, you wait
00:00:42.840 --> 00:00:44.880
a beat, and then you hear the boom. Cause and
00:00:44.880 --> 00:00:47.439
effect, right? Sound and action. You speak a
00:00:47.439 --> 00:00:49.240
command to your phone, it lights up, it does
00:00:49.240 --> 00:00:51.479
the thing. Because we live in the thunder. I
00:00:51.479 --> 00:00:53.200
mean, we've built literally all of our voice
00:00:53.200 --> 00:00:56.039
technology around waiting for that acoustic boom
00:00:56.039 --> 00:00:59.079
to hit a microphone. Exactly. But what if a computer
00:00:59.079 --> 00:01:01.119
just didn't need to wait for the thunder anymore?
00:01:01.320 --> 00:01:03.179
What if it could see the invisible lightning
00:01:03.179 --> 00:01:05.879
happening like right under your skin milliseconds
00:01:05.879 --> 00:01:09.000
before you actually make a sound? Okay, let's
00:01:09.000 --> 00:01:11.519
unpack this. Let's do it. We are looking at a
00:01:11.519 --> 00:01:14.560
massive, completely under -the -radar acquisition
00:01:14.560 --> 00:01:18.799
by Apple that went down in January 2026. They
00:01:18.799 --> 00:01:21.480
bought a company called Q .AI, formerly known
00:01:21.480 --> 00:01:26.340
as Q, or QLTD, for a reported $1 .5 to $2 billion.
00:01:26.840 --> 00:01:29.299
Which is just a staggering amount of money. Right.
00:01:29.359 --> 00:01:31.500
That makes it Apple's second largest acquisition
00:01:31.500 --> 00:01:34.420
in history, right behind Beats. And we're using
00:01:34.420 --> 00:01:37.000
Brian Rommel's exclusive analysis today to look
00:01:37.000 --> 00:01:39.379
past the mainstream media's incredibly superficial
00:01:39.379 --> 00:01:41.269
coverage. Ridge of the Steel. They really missed
00:01:41.269 --> 00:01:43.010
the mark on this one. I mean, completely. Oh,
00:01:43.049 --> 00:01:45.230
entirely. So our mission for you on this deep
00:01:45.230 --> 00:01:47.549
dive is to understand how this purchase isn't
00:01:47.549 --> 00:01:50.069
just an audio upgrade. You know, it represents
00:01:50.069 --> 00:01:51.989
a foundational leap towards something called
00:01:51.989 --> 00:01:54.450
connectionless intent reading. And we really
00:01:54.450 --> 00:01:56.769
need to emphasize the sheer scale of that number.
00:01:57.049 --> 00:02:00.739
Two. billion. That is not an acqui -hire. No,
00:02:00.799 --> 00:02:03.760
definitely not. That is not buying a neat little
00:02:03.760 --> 00:02:07.159
feature to patch a bug in iOS. It is a foundational
00:02:07.159 --> 00:02:10.620
bet on a completely new paradigm for human -computer
00:02:10.620 --> 00:02:14.500
interaction. It's bypassing the acoustic requirement
00:02:14.500 --> 00:02:17.439
of computing entirely. Right. But if you look
00:02:17.439 --> 00:02:19.699
at the public narrative right now, like there's
00:02:19.699 --> 00:02:22.180
this tech news carousel that's been blowing up
00:02:22.180 --> 00:02:25.539
on Instagram all week. The whole Q .AI purchase
00:02:25.539 --> 00:02:28.699
is being framed as a routine AI shopping trip.
00:02:28.819 --> 00:02:31.360
Oh, yeah. The whispered speech. Yeah. They're
00:02:31.360 --> 00:02:34.020
saying, oh, Apple spent a couple billion dollars
00:02:34.020 --> 00:02:37.699
for whispered speech technology. or noise -robust
00:02:37.699 --> 00:02:40.759
audio. The narrative is basically, well, Apple
00:02:40.759 --> 00:02:42.800
bought this, so Siri finally works when you're
00:02:42.800 --> 00:02:44.800
standing on a noisy subway platform. Which is
00:02:44.800 --> 00:02:47.659
hilarious. Even Johnny Sruji, you know, Apple's
00:02:47.659 --> 00:02:50.120
hardware chief, publicly called the Q .AI firm
00:02:50.120 --> 00:02:53.199
remarkable in this really vague, noncommittal
00:02:53.199 --> 00:02:55.360
way. Which is exactly what Apple wants everyone
00:02:55.360 --> 00:02:57.560
to think. Whispered speech is honestly the perfect
00:02:57.560 --> 00:02:59.419
cover story. Because it sounds boring. Right.
00:02:59.439 --> 00:03:01.780
It's a plausible, easily digestible narrative
00:03:01.780 --> 00:03:03.639
that they can say out loud without tipping their
00:03:03.639 --> 00:03:05.780
hand to competitors. But, I mean, think about
00:03:05.780 --> 00:03:07.780
it. out the economics for a second. Solving background
00:03:07.780 --> 00:03:10.840
noise on a subway is a $50 million software problem.
00:03:11.379 --> 00:03:13.960
You don't drop $2 billion to build a better noise
00:03:13.960 --> 00:03:16.840
gate. That's what I'm saying. Framing a $2 billion
00:03:16.840 --> 00:03:19.620
acquisition as a noise cancellation upgrade is
00:03:19.620 --> 00:03:22.699
like, I don't know, buying a billion dollar supercomputer
00:03:22.699 --> 00:03:25.000
just so you can use the calculator app to tip
00:03:25.000 --> 00:03:26.900
your waiter. Yeah, makes zero sense. It makes
00:03:26.900 --> 00:03:29.330
no sense at all. So if it's not a chatbot. And
00:03:29.330 --> 00:03:31.469
it's not noise cancellation. What did Apple actually
00:03:31.469 --> 00:03:34.330
buy? They bought a way to read words that are
00:03:34.330 --> 00:03:36.710
not fully spoken. Which honestly gives me chills
00:03:36.710 --> 00:03:39.710
just thinking about it. But to really grasp how
00:03:39.710 --> 00:03:42.430
you read an unspoken word, we can't just look
00:03:42.430 --> 00:03:45.930
at January 2026. I want to take a step back.
00:03:46.169 --> 00:03:48.389
We have to look at the history. Yeah, we actually
00:03:48.389 --> 00:03:50.830
have to travel back to trace a 13 -year circle
00:03:50.830 --> 00:03:54.830
that starts in 2013 with one very specific founder.
00:03:55.150 --> 00:03:57.750
Eviad Maisels. That's the guy. And this history
00:03:57.750 --> 00:04:00.580
isn't just trivia. You know, it is the absolute
00:04:00.580 --> 00:04:03.319
blueprint for what is happening today. Without
00:04:03.319 --> 00:04:05.879
understanding his track record, the Q .I deal
00:04:05.879 --> 00:04:08.500
looks like magic. But with his track record,
00:04:08.680 --> 00:04:11.039
it looks like inevitable physics. Let's lay out
00:04:11.039 --> 00:04:13.659
that track record for everyone. Back in November
00:04:13.659 --> 00:04:17.740
2013, Apple bought a Tel Aviv -based company
00:04:17.740 --> 00:04:22.480
called PrimeSense for about $360 million. A pretty
00:04:22.480 --> 00:04:24.779
big deal at the time. Yeah. Now, if you were
00:04:24.779 --> 00:04:27.029
paying attention to tech back then, You probably
00:04:27.029 --> 00:04:29.269
know PrimeSense as the company that built the
00:04:29.269 --> 00:04:31.689
DepthEye for Microsoft's original Kinect gaming
00:04:31.689 --> 00:04:35.050
system. Right. And at the time, the entire tech
00:04:35.050 --> 00:04:37.670
press got the narrative completely wrong, just
00:04:37.670 --> 00:04:39.750
like they are today with Q2II. Oh, they were
00:04:39.750 --> 00:04:42.129
so... Everyone thought Apple was going to release
00:04:42.129 --> 00:04:45.149
an Apple TV gaming console. They thought, oh,
00:04:45.230 --> 00:04:47.829
Apple bought a Kinect supplier to compete with
00:04:47.829 --> 00:04:50.660
Xbox. And then nothing. Absolute crickets for
00:04:50.660 --> 00:04:54.079
four years. From 2013 to 2017, Apple didn't say
00:04:54.079 --> 00:04:56.680
a word about PrimeSense, which, by the way, four
00:04:56.680 --> 00:04:58.800
years of silence in the tech world is an eternity.
00:04:58.980 --> 00:05:01.339
It's a lifetime. So what is happening inside
00:05:01.339 --> 00:05:03.720
Apple during a latency period like that? What
00:05:03.720 --> 00:05:05.860
are they doing with a gaming sensor? Well, you're
00:05:05.860 --> 00:05:07.779
taking a piece of hardware that was basically
00:05:07.779 --> 00:05:11.120
the size of a brick, the Kinect sensor, and you
00:05:11.120 --> 00:05:14.399
are brutalizing the supply chain to miniaturize
00:05:14.399 --> 00:05:16.850
it. Shrinking it down. Exactly. You're working
00:05:16.850 --> 00:05:19.949
out yield rates, thermal dynamics, and figuring
00:05:19.949 --> 00:05:23.649
out how to mass produce this exotic optical sensor
00:05:23.649 --> 00:05:26.250
so you can cram it into the top millimeter of
00:05:26.250 --> 00:05:28.649
a smartphone display for hundreds of millions
00:05:28.649 --> 00:05:30.910
of people. Right. And when it finally emerged
00:05:30.910 --> 00:05:34.009
from that black box in 2017, it wasn't a gaming
00:05:34.009 --> 00:05:37.470
peripheral at all. That exact same optical technology
00:05:37.470 --> 00:05:40.610
using near -infrared light to map a spaceship
00:05:40.610 --> 00:05:43.290
as the TrueDepth array. Face ID. We know it as
00:05:43.290 --> 00:05:46.259
Face ID today, yeah. Which is why... Brian Rommel's
00:05:46.259 --> 00:05:48.939
research from back then is so critical to look
00:05:48.939 --> 00:05:52.620
at today. Years before Face ID launched, he argued
00:05:52.620 --> 00:05:55.259
on Quora that PrimeSense was never a gaming leftover.
00:05:55.639 --> 00:05:57.959
He called it. He totally called it. He saw it
00:05:57.959 --> 00:06:00.160
as the first tile in a sensing stack that Apple
00:06:00.160 --> 00:06:02.980
was assembling for identity, emotion, and passwordless
00:06:02.980 --> 00:06:05.819
devices. He understood that Apple didn't buy
00:06:05.819 --> 00:06:07.660
a gaming product. Right, they bought the people.
00:06:07.939 --> 00:06:10.019
Exactly. They bought a team that had mastered
00:06:10.019 --> 00:06:12.540
a specific type of physics. And that brings us
00:06:12.540 --> 00:06:15.560
to the 2026 parallel, because the president and
00:06:15.560 --> 00:06:18.100
co -founder of a PrimeSense was Avi Edmazels.
00:06:18.279 --> 00:06:21.819
And in January 2026, this exact same guy, along
00:06:21.819 --> 00:06:24.540
with his co -founders Yonatan Wexler and Avi
00:06:24.540 --> 00:06:27.899
Barlia, sells a second company to Apple. QII.
00:06:28.180 --> 00:06:30.399
QII. They are going back inside Apple. Now, I
00:06:30.399 --> 00:06:32.660
have to stop here. Because Apple is one of the
00:06:32.660 --> 00:06:36.199
most resource -rich companies on Earth. Why drop
00:06:36.199 --> 00:06:39.740
$2 billion on the exact same founder? It's a
00:06:39.740 --> 00:06:42.459
great question. Like, is it nostalgia? Or do
00:06:42.459 --> 00:06:44.879
Maisel's and his team possess some totally unique
00:06:44.879 --> 00:06:47.079
approach to hardware that Apple's own tenured
00:06:47.079 --> 00:06:49.240
engineers just can't crack? Well, Apple could
00:06:49.240 --> 00:06:51.079
probably figure it out eventually, you know.
00:06:51.160 --> 00:06:53.939
But time is the one thing Apple can't manufacture.
00:06:54.220 --> 00:06:56.379
That makes sense. Maisel's team has a highly
00:06:56.379 --> 00:06:58.879
specialized, proven mastery over the physics
00:06:58.879 --> 00:07:01.889
of light interacting with human biology. What's
00:07:01.889 --> 00:07:04.029
fascinating here is, if you look at the lineage
00:07:04.029 --> 00:07:06.029
CrimeSense taught machines how to see the world
00:07:06.029 --> 00:07:08.750
as volume. As 3D space. Right. They used physics
00:07:08.750 --> 00:07:10.970
to solve a massive input problem, which was typing
00:07:10.970 --> 00:07:13.750
in a passcode or using a fingerprint. Face ID
00:07:13.750 --> 00:07:15.910
meant the machine finally understood your geometry.
00:07:16.230 --> 00:07:18.329
It stopped asking you to perform the physical
00:07:18.329 --> 00:07:20.889
action of pressing your thumb on a glass circle.
00:07:21.230 --> 00:07:23.470
Precisely. Now apply that exact same philosophy
00:07:23.470 --> 00:07:27.420
to Q .ai. This second company takes that mastery
00:07:27.420 --> 00:07:29.959
of optical physics and applies it to a completely
00:07:29.959 --> 00:07:32.259
different biological process. Okay, I'm tracking.
00:07:32.500 --> 00:07:34.839
Q .ai teaches machines to see the preparation
00:07:34.839 --> 00:07:37.879
of language. Just as PrimeSense eliminated the
00:07:37.879 --> 00:07:40.740
fingerprint, Q .ai provides the physics required
00:07:40.740 --> 00:07:44.120
to eliminate the wake word. Okay, so both companies
00:07:44.120 --> 00:07:47.319
solve computing input problems using light. FaceID
00:07:47.319 --> 00:07:49.620
gets rid of the fingerprint. Q .ai gets rid of
00:07:49.620 --> 00:07:51.899
HeySiri. But I want to really dig into the how
00:07:51.899 --> 00:07:54.649
here. We need to. The mechanics are crazy. Because
00:07:54.649 --> 00:07:56.410
the source material goes deep into the physics,
00:07:56.569 --> 00:07:59.350
and we need to contrast how Face ID works versus
00:07:59.350 --> 00:08:01.889
how Q .AI works. Let's start with Face ID, because
00:08:01.889 --> 00:08:03.310
most people have it in their pocket right now.
00:08:03.550 --> 00:08:06.149
Face ID uses something called structured light
00:08:06.149 --> 00:08:08.569
coding. Okay. When you look at your phone, a
00:08:08.569 --> 00:08:11.930
dot projector fires about 30 ,000 tiny dots of
00:08:11.930 --> 00:08:13.990
near -infrared light onto your face. And because
00:08:13.990 --> 00:08:16.750
my face isn't flat, I've got a nose, cheekbones,
00:08:16.829 --> 00:08:19.509
eye sockets. That 3D geometry warps the pattern
00:08:19.509 --> 00:08:22.649
of those 30 ,000 dots, right? Precisely. Your
00:08:22.649 --> 00:08:26.529
anatomy distorts the grid. Then a standard CMOS
00:08:26.529 --> 00:08:29.149
sensor, which is essentially just the basic digital
00:08:29.149 --> 00:08:31.870
camera chip that's in every smartphone, reads
00:08:31.870 --> 00:08:34.289
that distortion. So it's basically taking a picture
00:08:34.289 --> 00:08:37.289
of the bent dots. Yeah, and clever software calculates
00:08:37.289 --> 00:08:40.090
the depth based on how the dot's bent. But here's
00:08:40.090 --> 00:08:43.850
the key. Face ID maps your face as a rigid, static
00:08:43.850 --> 00:08:47.450
3D object. It's looking at the macroscopic structure
00:08:47.450 --> 00:08:50.159
of your head. I'm trying to visualize this. Let's
00:08:50.159 --> 00:08:53.139
say Face ID is like a satellite taking a high
00:08:53.139 --> 00:08:55.399
-resolution photograph of a frozen lake. Oh,
00:08:55.460 --> 00:08:57.580
that's a good analogy. It maps the surface perfectly.
00:08:57.759 --> 00:09:00.500
It sees the static shape of the ice, the ridges,
00:09:00.519 --> 00:09:03.139
the shoreline. It creates a perfect model of
00:09:03.139 --> 00:09:05.580
a large, rigid object. That's a great baseline.
00:09:05.700 --> 00:09:08.120
Now, Q .AI is not looking at the static ice.
00:09:08.299 --> 00:09:11.120
Right. Q .AI is the next resolution down. Much,
00:09:11.120 --> 00:09:14.360
much further down. Q .AI doesn't project 30 ,000
00:09:14.360 --> 00:09:16.840
dots over your whole face. It shines a very targeted
00:09:16.840 --> 00:09:19.480
beam of coherence. light onto a very small patch
00:09:19.480 --> 00:09:21.860
of your facial skin. Wait, pause. What makes
00:09:21.860 --> 00:09:24.320
light coherent? Why does that matter? Good question.
00:09:24.639 --> 00:09:27.259
Think of the light bulb in your lamp. It throws
00:09:27.259 --> 00:09:29.779
light out in every direction, totally disorganized.
00:09:29.879 --> 00:09:32.299
That's incoherent. Like just splashing water
00:09:32.299 --> 00:09:35.850
everywhere. Exactly. Coherent light, like a laser,
00:09:35.990 --> 00:09:38.169
consists of light waves that are marching in
00:09:38.169 --> 00:09:40.509
perfect lockstep with one another. They are highly
00:09:40.509 --> 00:09:42.909
organized. Okay, got it. When you shine coherent
00:09:42.909 --> 00:09:45.750
light onto a surface like human skin and it bounces
00:09:45.750 --> 00:09:48.429
back to a detector, it creates a very specific
00:09:48.429 --> 00:09:51.789
interference pattern called a speckle. A speckle.
00:09:51.889 --> 00:09:54.669
Okay, so the coherent light bounces off my skin
00:09:54.669 --> 00:09:56.929
and creates this speckle pattern on the sensor.
00:09:57.330 --> 00:09:59.850
Where does the speech part come in? This is where
00:09:59.850 --> 00:10:02.409
we cross from physics into biology. Buckle up.
00:10:02.610 --> 00:10:04.950
The source analysis explains that when you are
00:10:04.950 --> 00:10:07.409
about to speak before air ever leaves your lungs,
00:10:07.610 --> 00:10:10.169
the muscles that form a vowel or a consonant
00:10:10.169 --> 00:10:13.230
begin to twitch. But these are tiny. Right? Oh,
00:10:13.409 --> 00:10:15.269
we're talking about twitches on the scale of
00:10:15.269 --> 00:10:18.350
tens of microns. A micron is one thousandth of
00:10:18.350 --> 00:10:20.909
a millimeter. It is completely invisible to the
00:10:20.909 --> 00:10:23.090
human eye. But not invisible to the coherent
00:10:23.090 --> 00:10:26.470
light. Exactly. When that microscopic muscle
00:10:26.470 --> 00:10:29.090
twitch happens under your skin, it changes the
00:10:29.090 --> 00:10:31.549
surface of your skin by a fraction of hair's
00:10:31.549 --> 00:10:34.710
width. But because the light is coherent, that
00:10:34.710 --> 00:10:37.549
tiny shift violently disrupts the speckle pattern
00:10:37.549 --> 00:10:39.950
bouncing back to the sensor. Whoa! The sensor
00:10:39.950 --> 00:10:43.009
reads that shifting speckle. Then... A highly
00:10:43.009 --> 00:10:46.190
trained neural network maps that specific speckle
00:10:46.190 --> 00:10:49.350
shift onto phonemes. And phonemes are just the
00:10:49.350 --> 00:10:51.409
distinct sounds that make up a language, like
00:10:51.409 --> 00:10:54.110
the K sound in cat or the B sound in bat. Right.
00:10:54.250 --> 00:10:56.590
The neural network pieces those phones together
00:10:56.590 --> 00:10:59.389
into words and ultimately into your intention
00:10:59.389 --> 00:11:02.129
to speak before the speech has ever become an
00:11:02.129 --> 00:11:04.669
audible sound. That is mind -bending. It is reading
00:11:04.669 --> 00:11:07.330
the biological motor plan of language. And it
00:11:07.330 --> 00:11:09.570
does this entirely optically, from a distance.
00:11:10.029 --> 00:11:12.929
No electrodes taped to your throat. No physical
00:11:12.929 --> 00:11:15.509
contact with your skin. So going back to my frozen
00:11:15.509 --> 00:11:18.570
lake analogy, if Face ID is the satellite mapping
00:11:18.570 --> 00:11:20.590
the frozen static surface of the ice, then Q
00:11:20.590 --> 00:11:23.149
.AI is like shining a laser onto the water before
00:11:23.149 --> 00:11:25.649
it freezes, watching the microscopic ripples
00:11:25.649 --> 00:11:27.909
and currents beneath the surface to predict exactly
00:11:27.909 --> 00:11:30.710
where and how a wave is going to break. It's
00:11:30.710 --> 00:11:32.990
reading the energy before the event happens.
00:11:33.210 --> 00:11:35.629
That captures the physics perfectly. Yeah. And
00:11:35.629 --> 00:11:38.409
this physical leap is staggering. It moves the
00:11:38.409 --> 00:11:41.730
entire computing industry from reading static
00:11:41.730 --> 00:11:45.279
geometry anatomy to... reading living dynamic
00:11:45.279 --> 00:11:48.700
biological processes in real time. We are shifting
00:11:48.700 --> 00:11:51.980
to physiology. Exactly. And this isn't just some
00:11:51.980 --> 00:11:54.600
theoretical white paper either. According to
00:11:54.600 --> 00:11:57.419
Rommel's analysis, Apple has already mapped this
00:11:57.419 --> 00:12:00.399
out meticulously in a vast web of patents that
00:12:00.399 --> 00:12:03.340
they now own. They have the blueprint. And that
00:12:03.340 --> 00:12:06.039
blueprint is how you justify a $2 billion price
00:12:06.039 --> 00:12:10.059
tag. The analysis points out that Q .AI wasn't
00:12:10.059 --> 00:12:12.940
just a single parlor trick. It is an entire product
00:12:12.940 --> 00:12:15.700
architecture written in patent claims. Right.
00:12:15.799 --> 00:12:17.440
It's not just a sensor. It's a whole system.
00:12:17.580 --> 00:12:19.379
When you strip away the patent numbers and just
00:12:19.379 --> 00:12:22.179
look at what these documents describe, a massive,
00:12:22.320 --> 00:12:25.320
highly disruptive product roadmap emerges. Let's
00:12:25.320 --> 00:12:26.779
walk through the architecture because I don't
00:12:26.779 --> 00:12:28.539
want to just list off a bunch of eight -digit
00:12:28.539 --> 00:12:30.220
patent registry numbers that nobody's going to
00:12:30.220 --> 00:12:32.120
remember. Let's look at what these patents actually
00:12:32.120 --> 00:12:34.120
do. do. Yeah, the practical application. The
00:12:34.120 --> 00:12:36.320
first one we have to talk about is what I'll
00:12:36.320 --> 00:12:38.840
call the skin transcription patent. This is the
00:12:38.840 --> 00:12:42.450
foundational one. The claim describes deciphering
00:12:42.450 --> 00:12:45.250
silent speech by shining coherent light onto
00:12:45.250 --> 00:12:48.029
the skin, reading the speckle coming in, and
00:12:48.029 --> 00:12:50.769
generating transcribed language out. It's basically
00:12:50.769 --> 00:12:52.950
the core mechanic we just talked about. Exactly.
00:12:52.950 --> 00:12:56.029
It explicitly covers people articulating words
00:12:56.029 --> 00:12:58.370
without making any physical contact with the
00:12:58.370 --> 00:13:00.710
sensor. Okay, so that's the core physics. But
00:13:00.710 --> 00:13:02.950
then the next group of patents answers the obvious
00:13:02.950 --> 00:13:05.190
question, where does this laser actually live
00:13:05.190 --> 00:13:08.830
on my body? And that brings us to the earbud
00:13:08.830 --> 00:13:11.169
housing patent. This is where it gets real for
00:13:11.169 --> 00:13:13.740
the consumer. The patents detail determining
00:13:13.740 --> 00:13:16.799
speech from skin movements using a housing supported
00:13:16.799 --> 00:13:20.360
by the human ear. So AirPods. This is the AirPods
00:13:20.360 --> 00:13:23.320
claim set. It tells us exactly what hardware
00:13:23.320 --> 00:13:25.860
ecosystem this sensor is destined for. Which
00:13:25.860 --> 00:13:28.279
makes perfect sense physically. The stem of an
00:13:28.279 --> 00:13:30.940
AirPod points right down at your cheek and jaw
00:13:30.940 --> 00:13:33.460
hinge. It's in the perfect position to shine
00:13:33.460 --> 00:13:35.440
that coherent light. It couldn't be positioned
00:13:35.440 --> 00:13:37.840
better. But here is the patent that absolutely
00:13:37.840 --> 00:13:40.679
blew my mind. Let's call it the intent patent.
00:13:41.210 --> 00:13:44.590
Oh, this one is wild. It's literally titled interpreting
00:13:44.590 --> 00:13:48.190
words prior to vocalization alongside another
00:13:48.190 --> 00:13:52.309
for facial movements, waking up a wearable. Interpreting
00:13:52.309 --> 00:13:55.409
words prior to vocalization. That is the death
00:13:55.409 --> 00:13:58.009
of the wake word right there. The early muscular
00:13:58.009 --> 00:14:00.110
recruitment of your face getting ready to speak
00:14:00.110 --> 00:14:03.490
is the new. Haste, Siri. Think about how profound
00:14:03.490 --> 00:14:06.049
that is. You don't announce your intention to
00:14:06.049 --> 00:14:08.529
a machine anymore. The machine observes your
00:14:08.529 --> 00:14:10.889
biology, preparing to act. And they pair that
00:14:10.889 --> 00:14:13.450
with what we should call the privacy output patent.
00:14:13.990 --> 00:14:16.250
Because if a machine can read your silent query,
00:14:16.549 --> 00:14:18.769
it needs to give you a silent answer. Right.
00:14:18.830 --> 00:14:20.470
You don't want it shouting the answer out loud.
00:14:20.840 --> 00:14:23.059
This patent covers providing private answers
00:14:23.059 --> 00:14:26.460
to non -vocal questions, routing the audio discreetly
00:14:26.460 --> 00:14:28.620
back into your earbud rather than blasting it
00:14:28.620 --> 00:14:30.659
out of a phone speaker. Okay, I have to jump
00:14:30.659 --> 00:14:33.139
in here with a very practical, real -world concern
00:14:33.139 --> 00:14:35.559
based on the source text. Sure, go ahead. If
00:14:35.559 --> 00:14:38.100
I have this future AirPod in my ear and it is
00:14:38.100 --> 00:14:40.580
constantly reading the microscopic muscle twitches
00:14:40.580 --> 00:14:42.799
of my cheek and jaw, what happens if I'm just
00:14:42.799 --> 00:14:45.279
chewing a piece of gum? Or eating a sandwich?
00:14:45.500 --> 00:14:47.759
Or what if I smile at someone on the street or
00:14:47.759 --> 00:14:49.679
just have a random nervous twitch? This is the
00:14:49.679 --> 00:14:52.000
first thing everyone asks. Is my phone going
00:14:52.000 --> 00:14:54.720
to constantly misinterpret that as a word and
00:14:54.720 --> 00:14:57.879
fire off gibberish texts to my boss? That is
00:14:57.879 --> 00:15:00.820
the exact hurdle that separates a neat lab experiment
00:15:00.820 --> 00:15:04.320
from a shipping consumer product. But the patent
00:15:04.320 --> 00:15:06.740
portfolio accounts for this. They have a whole
00:15:06.740 --> 00:15:09.500
set of claims covering threshold intensity. What
00:15:09.500 --> 00:15:11.769
does that mean in plain English? Threshold intensity.
00:15:12.169 --> 00:15:13.970
It means the neural network isn't just looking
00:15:13.970 --> 00:15:17.090
for any movement. Chewing gum creates a massive
00:15:17.090 --> 00:15:20.750
rhythmic macro movement of the jaw. Smiling stretches
00:15:20.750 --> 00:15:23.990
the skin in a very specific global way. So the
00:15:23.990 --> 00:15:25.830
sensor can tell the difference. The system is
00:15:25.830 --> 00:15:27.830
trained to filter those out as optical noise.
00:15:28.490 --> 00:15:30.830
It is looking for the highly specific, incredibly
00:15:30.830 --> 00:15:33.710
subtle micro sequencing of a motor plan forming
00:15:33.710 --> 00:15:36.610
a phone. It knows the difference between a chew
00:15:36.610 --> 00:15:39.090
and a consonant. So when you synthesize all of
00:15:39.090 --> 00:15:42.070
these pieces. The skin transcription, the earbud
00:15:42.070 --> 00:15:45.269
housing, the pre -vocalization intent, the privacy
00:15:45.269 --> 00:15:48.029
output, the threshold filters. You realize Apple
00:15:48.029 --> 00:15:50.370
didn't buy a microphone company. Not at all.
00:15:50.470 --> 00:15:53.570
They bought an entirely new input layer. It's
00:15:53.570 --> 00:15:55.889
a computer interface that no longer requires
00:15:55.889 --> 00:15:58.970
the physical act of spoken speech. It bypasses
00:15:58.970 --> 00:16:02.009
the air entirely. But having the patents in a
00:16:02.009 --> 00:16:04.509
vault in Cupertino is one thing. How does this
00:16:04.509 --> 00:16:07.429
actually integrate into the Apple devices that
00:16:07.429 --> 00:16:09.649
you, the listener, are carrying in your pocket?
00:16:10.299 --> 00:16:16.320
That's the billion -dollar question. I want to
00:16:16.320 --> 00:16:18.340
pause on the word connectionless because we aren't
00:16:18.340 --> 00:16:20.399
talking about losing your Wi -Fi signal here.
00:16:20.539 --> 00:16:22.779
No, connectionless here refers to the user experience.
00:16:23.320 --> 00:16:25.820
Think about how every voice assistant has functioned
00:16:25.820 --> 00:16:28.899
since Siri launched in 2011. You have to establish
00:16:28.899 --> 00:16:30.639
a connection. You have to initiate it. Right.
00:16:30.740 --> 00:16:33.580
You shout a wake word across a room, or you unlock
00:16:33.580 --> 00:16:35.899
your phone, open an app, and hold down a button.
00:16:36.570 --> 00:16:39.389
You perform an action to open an acoustic session.
00:16:39.590 --> 00:16:42.350
But with this tech, I don't have to pair anything
00:16:42.350 --> 00:16:45.070
new. I don't hold a button. I don't even have
00:16:45.070 --> 00:16:47.389
to finish the word I'm thinking of. The device
00:16:47.389 --> 00:16:50.490
is already optically coupled to my face. To truly
00:16:50.490 --> 00:16:53.429
understand how revolutionary this is, the source
00:16:53.429 --> 00:16:55.610
material breaks down something called the speech
00:16:55.610 --> 00:16:58.799
cascade. I think this is crucial because most
00:16:58.799 --> 00:17:01.259
people misunderstand what speech actually is.
00:17:01.379 --> 00:17:03.960
Right. This totally flipped my perspective. We
00:17:03.960 --> 00:17:07.440
think of speech as a sound event. But sound is
00:17:07.440 --> 00:17:10.460
literally just the final exhaust of a much longer
00:17:10.460 --> 00:17:13.480
process. The very end of the line. The text outlines
00:17:13.480 --> 00:17:16.539
the cascade like this. First, there's the intent,
00:17:16.839 --> 00:17:19.380
the thought in your brain. Second, your brain
00:17:19.380 --> 00:17:22.460
creates a motor planning signal. Third, there
00:17:22.460 --> 00:17:24.759
is the recruitment of your facial and laryngeal
00:17:24.759 --> 00:17:27.579
muscles to execute that plan. And finally, fourth,
00:17:27.779 --> 00:17:30.039
air gets pushed from your lungs through those
00:17:30.039 --> 00:17:32.339
muscles to create sound. Old voice technology
00:17:32.339 --> 00:17:34.920
only interacts with step four, the air. Here's
00:17:34.920 --> 00:17:36.700
where it gets really interesting. Let me use
00:17:36.700 --> 00:17:38.740
another analogy. Think about a baseball game.
00:17:38.880 --> 00:17:41.500
The voice tech we use today only reacts once
00:17:41.500 --> 00:17:43.539
the ball has left the pitcher's hand and is flying
00:17:43.539 --> 00:17:46.029
through the air toward home plate. The microphone
00:17:46.029 --> 00:17:48.089
has to wait to catch the sound wave in flight.
00:17:48.250 --> 00:17:51.809
Sure. But QAI is reading the pitcher's windup.
00:17:51.950 --> 00:17:54.089
It reads the muscles in the pitcher's shoulder
00:17:54.089 --> 00:17:56.789
before the ball is ever released. It knows exactly
00:17:56.789 --> 00:17:59.049
what pitch is coming, how fast it'll be, and
00:17:59.049 --> 00:18:02.609
where it's aimed. That's spot on. Q .AI positions
00:18:02.609 --> 00:18:06.690
its sensors at step three of the cascade. It
00:18:06.690 --> 00:18:09.549
is, by definition, a brain -computer interface.
00:18:10.200 --> 00:18:14.059
A BCI. Wait, a BCI? Like Neuralink? But nobody
00:18:14.059 --> 00:18:16.200
is drilling holes in my head for this. Exactly.
00:18:16.400 --> 00:18:20.059
It's a non -surgical, non -invasive BCI. It reads
00:18:20.059 --> 00:18:22.299
the biological exhaust of your brain's motor
00:18:22.299 --> 00:18:24.480
plan without ever breaking the skin or forcing
00:18:24.480 --> 00:18:26.720
you to wear a skull cap of electrodes. That is
00:18:26.720 --> 00:18:29.660
wild. And when you map this onto Apple's current
00:18:29.660 --> 00:18:31.940
hardware stack, the elegance of the integration
00:18:31.940 --> 00:18:34.700
is undeniable. If we connect this to the bigger
00:18:34.700 --> 00:18:37.150
picture, let's start with the AirPods. We talked
00:18:37.150 --> 00:18:39.269
about the earbud patent earlier. The stem of
00:18:39.269 --> 00:18:41.150
the AirPod points right at the jaw. But the text
00:18:41.150 --> 00:18:43.670
mentions the earbud acting as a local interferometer.
00:18:43.809 --> 00:18:45.329
You're going to have to explain that one. An
00:18:45.329 --> 00:18:47.329
interferometer is just an instrument that bounces
00:18:47.329 --> 00:18:49.769
light waves off something and measures the interference
00:18:49.769 --> 00:18:51.809
pattern when the light comes back. Okay, so it
00:18:51.809 --> 00:18:53.990
measures interference. It's a way to measure
00:18:53.990 --> 00:18:56.750
incredibly tiny changes in distance or surface.
00:18:57.410 --> 00:19:00.670
So imagine the next generation of AirPods has
00:19:00.670 --> 00:19:03.549
a microscopic coherent light emitter and detector
00:19:03.549 --> 00:19:07.250
built into the plastic of the stem. Your AirPod
00:19:07.250 --> 00:19:09.630
is now an optical interferometer, constantly
00:19:09.630 --> 00:19:11.710
measuring the speckle on your cheek. Okay, so
00:19:11.710 --> 00:19:14.829
the AirPod is the primary sensor. But Apple's
00:19:14.829 --> 00:19:17.210
whole philosophy is ecosystem redundancy, right?
00:19:17.349 --> 00:19:19.650
The devices talk to each other. Right, which
00:19:19.650 --> 00:19:22.190
brings in the iPhone's TrueDepth camera, the
00:19:22.190 --> 00:19:24.990
Face ID sensor we talked about earlier. Oh, circling
00:19:24.990 --> 00:19:27.809
back to PrimeSense. Exactly. PrimeSense gave
00:19:27.809 --> 00:19:30.509
Apple the ability to throw structured light to
00:19:30.509 --> 00:19:33.329
verify your identity. QDAT AI gives them the
00:19:33.329 --> 00:19:35.349
knowledge to throw coherent light to read your
00:19:35.349 --> 00:19:37.509
language. Imagine you are holding your phone
00:19:37.509 --> 00:19:39.589
up, looking at the screen. Like scrolling or
00:19:39.589 --> 00:19:42.190
whatever. Right. The camera in the dynamic island
00:19:42.190 --> 00:19:45.009
can act as a secondary sensor, reading the speckle
00:19:45.009 --> 00:19:46.869
on your lips and face while the AirPod reads
00:19:46.869 --> 00:19:50.410
your cheek. Two optical opinions beat one. It
00:19:50.410 --> 00:19:53.569
helps disambiguate silent phonemes, but... All
00:19:53.569 --> 00:19:56.130
of this biological data, these constant speckle
00:19:56.130 --> 00:19:58.970
shifts, have to be processed instantly. Milliseconds.
00:19:58.990 --> 00:20:01.609
And this is where Apple's obsession with proprietary
00:20:01.609 --> 00:20:05.089
silicon comes in. The neural engine. This is
00:20:05.089 --> 00:20:07.490
the linchpin of the whole operation. Yeah. The
00:20:07.490 --> 00:20:09.789
patent descriptions specifically note that these
00:20:09.789 --> 00:20:12.650
silent speech models are highly optimized and
00:20:12.650 --> 00:20:14.970
small enough to run locally on the device's neural
00:20:14.970 --> 00:20:17.140
processing unit. Why does running it locally
00:20:17.140 --> 00:20:19.339
matter so much? Why not just send the speckled
00:20:19.339 --> 00:20:21.819
data to the cloud like ChatGPT does with text?
00:20:22.079 --> 00:20:24.900
Because of latency and privacy. If you have to
00:20:24.900 --> 00:20:27.380
beam microscopic muscle data to a server in Virginia
00:20:27.380 --> 00:20:29.640
to figure out what word you're mouthing, the
00:20:29.640 --> 00:20:31.740
delay breaks the magic. You'd be waiting for
00:20:31.740 --> 00:20:34.440
your own thought to load. Exactly. And more importantly,
00:20:34.599 --> 00:20:37.680
this perfectly aligns with Apple's 2026 on -device
00:20:37.680 --> 00:20:40.900
-first AI posture. You simply cannot sell a consumer
00:20:40.900 --> 00:20:43.700
a device that constantly reads their pre -vocalized
00:20:43.700 --> 00:20:46.220
biological intent if that data leaves the hardware.
00:20:46.420 --> 00:20:49.259
It has to be an absolute closed loop. And we
00:20:49.259 --> 00:20:51.980
can't forget about Vision Pro or whatever smart
00:20:51.980 --> 00:20:55.250
glasses Apple inevitably releases. A headset
00:20:55.250 --> 00:20:57.509
is inherently locked onto your face. Oh, the
00:20:57.509 --> 00:20:59.990
wearables aspect is huge. The source notes that
00:20:59.990 --> 00:21:02.430
Apple is already predicting user clicks based
00:21:02.430 --> 00:21:05.529
on pupil movement. Adding a cheek reading channel
00:21:05.529 --> 00:21:07.650
to a pair of smart glasses is just geometry.
00:21:08.170 --> 00:21:11.130
Pupil prediction was the dress rehearsal reading
00:21:11.130 --> 00:21:14.150
pre -speech intent is the main event. And beyond
00:21:14.150 --> 00:21:16.470
just convenience, think about the security implications.
00:21:16.890 --> 00:21:20.190
This entire stack creates a continuous authentication
00:21:20.190 --> 00:21:23.579
layer. How so? Well, Face ID occasionally asks,
00:21:23.700 --> 00:21:26.700
is this the correct rigid sculpture of a face?
00:21:27.180 --> 00:21:31.000
But a QII -enabled device constantly asks, is
00:21:31.000 --> 00:21:33.920
this the correct living dynamic muscle signature
00:21:33.920 --> 00:21:36.660
at this exact millisecond? It's the ultimate
00:21:36.660 --> 00:21:39.160
proof of liveness. Spoofing an optical speckle
00:21:39.160 --> 00:21:41.599
of a human muscle twitching in real time is exponentially
00:21:41.599 --> 00:21:44.319
harder than holding up a 3D -printed mask. Exactly.
00:21:44.730 --> 00:21:46.650
But let's pivot to the human element of this
00:21:46.650 --> 00:21:48.230
because this incredible hardware integration
00:21:48.230 --> 00:21:51.130
solves a massive behavioral problem that has
00:21:51.130 --> 00:21:53.509
played voice tech for over a decade. Right. The
00:21:53.509 --> 00:21:55.130
source analysis calls it the end of theater.
00:21:55.390 --> 00:21:58.289
Yes, because let's be honest with ourselves.
00:21:58.549 --> 00:22:01.109
The reason voice assistants like Siri, Alexa,
00:22:01.349 --> 00:22:03.869
or Google haven't become our primary computing
00:22:03.869 --> 00:22:06.710
interface isn't just because they sometimes misunderstand
00:22:06.710 --> 00:22:09.690
us. No, the accuracy is actually pretty good
00:22:09.690 --> 00:22:12.150
now. The real failure was the social performance
00:22:12.150 --> 00:22:14.970
required to use them. The social friction is
00:22:14.970 --> 00:22:18.009
immense. People simply will not sit in a quiet
00:22:18.009 --> 00:22:20.569
open plan office and say, hey, Siri, summarize
00:22:20.569 --> 00:22:22.730
the quarterly financials. Or they won't stand
00:22:22.730 --> 00:22:25.130
on a crowded train and say, hey, Siri, text my
00:22:25.130 --> 00:22:27.529
doctor about my prescription. Right. The theater
00:22:27.529 --> 00:22:31.130
of voice computing is inherently awkward. Deeply
00:22:31.130 --> 00:22:33.569
unprivate. But with connectionless intent reading,
00:22:33.869 --> 00:22:36.630
users will transition to simply mouthing their
00:22:36.630 --> 00:22:39.450
queries. You don't whisper. You don't push air.
00:22:39.730 --> 00:22:41.549
You just move the muscles. You just let your
00:22:41.549 --> 00:22:43.970
facial muscles form the sentence silently and
00:22:43.970 --> 00:22:46.109
the earbud catches the motor plan and confirms
00:22:46.109 --> 00:22:49.500
the action privately in your ear. Remillel argues
00:22:49.500 --> 00:22:51.539
in the text that the requirement for our thoughts
00:22:51.539 --> 00:22:53.740
to become physical air pressure before a computer
00:22:53.740 --> 00:22:56.259
could understand them was always a bug in computing.
00:22:56.559 --> 00:22:58.380
We just accepted it because we didn't have the
00:22:58.380 --> 00:23:01.099
physics to bypass it yet. It's a profound observation.
00:23:01.579 --> 00:23:04.200
However, as analysts, we have a responsibility
00:23:04.200 --> 00:23:07.359
to look at the real -world caveats here. The
00:23:07.359 --> 00:23:10.059
distance between a pristine patent filing and
00:23:10.059 --> 00:23:12.980
a shipping consumer product is a minefield. Oh,
00:23:13.019 --> 00:23:15.490
it's a massive gap. We have to do a reality check.
00:23:15.670 --> 00:23:17.990
Right, let's do a reality check. The analysis
00:23:17.990 --> 00:23:20.609
outlines some very real constraints. First off,
00:23:20.730 --> 00:23:23.750
historically, silent speech interfaces built
00:23:23.750 --> 00:23:26.509
in university labs have totally collapsed outside
00:23:26.509 --> 00:23:28.690
of speaker -specific training. They're incredibly
00:23:28.690 --> 00:23:31.750
fragile. Like, they work flawlessly for the one
00:23:31.750 --> 00:23:34.589
grad student who spent 100 hours calibrating
00:23:34.589 --> 00:23:37.549
the sensor to their specific jaw structure, but
00:23:37.549 --> 00:23:39.869
they fail the moment someone else puts it on.
00:23:39.990 --> 00:23:42.619
And that's in a lab. Reading an optical speckle
00:23:42.619 --> 00:23:44.880
on a cheek in a perfectly lit, climate -controlled
00:23:44.880 --> 00:23:47.319
environment is one thing. Trying to read that
00:23:47.319 --> 00:23:49.559
same speckle on a sweaty runner's cheek while
00:23:49.559 --> 00:23:51.599
they are jogging down a windy street corner in
00:23:51.599 --> 00:23:53.839
Chicago is an entirely different physics problem.
00:23:54.039 --> 00:23:56.980
The environmental noise is massive. Wind, sweat,
00:23:57.240 --> 00:24:00.299
movement. Plus, the financial analysts tracking
00:24:00.299 --> 00:24:02.740
this deal still disagree on the final acquisition
00:24:02.740 --> 00:24:06.349
price by up to half a billion dollars. Which
00:24:06.349 --> 00:24:08.789
just shows how deeply under wraps this technology
00:24:08.789 --> 00:24:12.130
still is. Apple hasn't officially confirmed exactly
00:24:12.130 --> 00:24:15.650
how or when this will ship. Which brings us back
00:24:15.650 --> 00:24:18.069
to the PrimeSense timeline. Remember, that took
00:24:18.069 --> 00:24:20.390
four years of silence before Face ID shipped.
00:24:20.549 --> 00:24:23.130
Right. The engineering required to miniaturize
00:24:23.130 --> 00:24:26.109
a coherent light interferometer into an AirPod
00:24:26.109 --> 00:24:29.170
stem while maintaining battery life is monumental.
00:24:29.450 --> 00:24:32.769
If the deal closed in 2026, we shouldn't be surprised
00:24:32.769 --> 00:24:34.990
if we don't see this fully integrated until 2030.
00:24:35.529 --> 00:24:38.630
So as we look toward that 2030 horizon, what
00:24:38.630 --> 00:24:40.450
does this actually mean for us? I have to be
00:24:40.450 --> 00:24:42.829
honest, it sets off some massive privacy alarms
00:24:42.829 --> 00:24:45.630
for me. Understandably so. A sensor sitting on
00:24:45.630 --> 00:24:48.210
my face, constantly reading my pre -speech. Is
00:24:48.210 --> 00:24:50.170
this the ultimate helpful assistant that knows
00:24:50.170 --> 00:24:52.549
what I need with zero friction? Or is this the
00:24:52.549 --> 00:24:54.670
ultimate surveillance device if that invisible
00:24:54.670 --> 00:24:57.210
laser stays on when I don't want it to? It is
00:24:57.210 --> 00:25:00.210
entirely valid to feel uneasy about it. A sensor
00:25:00.210 --> 00:25:02.490
capable of reading intent before vocalization
00:25:02.490 --> 00:25:05.089
is one of the most potent technologies ever conceived.
00:25:05.329 --> 00:25:07.630
Right. Whether it is a tool or a surveillance
00:25:07.630 --> 00:25:09.829
monitor depends entirely on the architecture.
00:25:10.329 --> 00:25:13.349
Does Apple hold the model locally? Does the data
00:25:13.349 --> 00:25:15.849
stay on the neural engine? Can the user physically
00:25:15.849 --> 00:25:18.170
sever the optical connection? This is why that
00:25:18.170 --> 00:25:20.150
local processing we talked about earlier is the
00:25:20.150 --> 00:25:22.589
absolute linchpin. If this data went to the cloud,
00:25:22.650 --> 00:25:24.569
it would be a non -starter. Society would reject
00:25:24.569 --> 00:25:27.859
it. Precisely. The local processing is the only
00:25:27.859 --> 00:25:30.059
thing that gives this technology a chance to
00:25:30.059 --> 00:25:33.799
be accepted as a secure, empowering tool. But
00:25:33.799 --> 00:25:36.960
the reality remains, Apple now owns both the
00:25:36.960 --> 00:25:39.660
camera that proved you were you and the camera
00:25:39.660 --> 00:25:41.279
that can hear the sentence, you haven't even
00:25:41.279 --> 00:25:43.680
spoken yet. Wow. They own the identity layer,
00:25:43.819 --> 00:25:46.319
and now they own the intent layer. Back in 2013,
00:25:46.599 --> 00:25:48.460
the entire industry wondered why Apple would
00:25:48.460 --> 00:25:50.660
pay hundreds of millions of dollars for a Kinect
00:25:50.660 --> 00:25:53.400
gaming supplier. Four years later, the answer
00:25:53.400 --> 00:25:56.119
was sitting on the front of every iPhone in the
00:25:56.119 --> 00:25:58.819
world. And in 2026, the question is why Apple
00:25:58.819 --> 00:26:01.279
would pay $2 billion for a company with no consumer
00:26:01.279 --> 00:26:04.059
product and a laser aimed at a cheek. And the
00:26:04.059 --> 00:26:07.180
answer is already sitting in your ear. The AirPods
00:26:07.180 --> 00:26:09.619
are the connection. The intent is already written
00:26:09.619 --> 00:26:12.059
on your face. The only thing left for Apple to
00:26:12.059 --> 00:26:14.569
remove is the requirement that you speak. That's
00:26:14.569 --> 00:26:17.809
exactly it. Which leaves me with one final kind
00:26:17.809 --> 00:26:20.269
of provocative thought for you to chew on after
00:26:20.269 --> 00:26:23.470
this deep dive is over. We spent a lot of time
00:26:23.470 --> 00:26:26.130
talking about the speech cascade today, intent,
00:26:26.569 --> 00:26:29.410
motor planning, muscle recruitment, and finally,
00:26:29.549 --> 00:26:33.230
air. But what happens to human self -correction?
00:26:33.549 --> 00:26:35.829
Oh, that's an interesting angle. I mean, we have
00:26:35.829 --> 00:26:38.190
all had those moments where our brain fully intends
00:26:38.190 --> 00:26:40.750
to say something. Maybe you're angry. Your jaw
00:26:40.750 --> 00:26:43.529
muscles literally begin to form the words. But
00:26:43.529 --> 00:26:45.930
your social filter kicks in at the very last
00:26:45.930 --> 00:26:48.589
possible millisecond, and you bite your tongue.
00:26:48.730 --> 00:26:50.650
You stop yourself before making a sound. You
00:26:50.650 --> 00:26:53.710
hold back the thunder. It's true. We rely heavily
00:26:53.710 --> 00:26:55.890
on that acoustic delay to save us from our worst
00:26:55.890 --> 00:26:58.700
impulses. Exactly. So if Apple's new technology
00:26:58.700 --> 00:27:01.460
acts on our pre -vocalization, if it reads that
00:27:01.460 --> 00:27:03.900
microscopic muscle twitch that happens before
00:27:03.900 --> 00:27:06.900
our filter catches it, will our devices execute
00:27:06.900 --> 00:27:09.980
commands or send private messages that we actually
00:27:09.980 --> 00:27:11.980
changed our minds about a fraction of a second
00:27:11.980 --> 00:27:14.200
later? It's a terrifying thought. What happens
00:27:14.200 --> 00:27:17.220
to us socially and psychologically when the gap
00:27:17.220 --> 00:27:19.960
between impulse and action is completely erased?
00:27:20.259 --> 00:27:21.920
It makes you wonder if living in the thunder
00:27:21.920 --> 00:27:24.380
wasn't so bad after all. Thank you for joining
00:27:24.380 --> 00:27:26.279
us on this deep dive into the source material.
00:27:26.500 --> 00:27:45.150
We'll catch you on the next one. Technology and
00:27:45.150 --> 00:27:49.750
history, he connects them all for you Brian Romer
00:27:49.750 --> 00:27:52.049
leads the way with research deep and true Denology
00:27:52.049 --> 00:27:54.230
and history, he connects them all for you Ancient
00:27:54.230 --> 00:27:56.730
sparks to modern code, he traces every thread
00:28:29.519 --> 00:28:34.079
But humanity moves on. Love equation for AI.
00:28:34.660 --> 00:28:39.299
Zero human companies building from the start.
00:28:40.579 --> 00:28:44.680
Thermodynamic wages earning the key. Secrets
00:28:44.680 --> 00:28:45.640
we will read.
