WEBVTT

00:00:00.000 --> 00:00:03.799
You don't get to an $852 billion valuation without

00:00:03.799 --> 00:00:07.780
breaking a few eggs. Beat. But what if the CEO

00:00:07.780 --> 00:00:10.740
is breaking the entire kitchen? Uh, yeah, that

00:00:10.740 --> 00:00:13.000
is the big question right now. Welcome to the

00:00:13.000 --> 00:00:16.019
deep dive. We are unpacking a massive, unprecedented

00:00:16.019 --> 00:00:20.179
leak of internal OpenAI documents today. And

00:00:20.179 --> 00:00:22.699
we are tracking massive industry shifts from

00:00:22.699 --> 00:00:25.600
Elon's sprawling lawsuit to Google's privacy

00:00:25.600 --> 00:00:28.859
invasion. Right. Then we explore entirely new

00:00:28.859 --> 00:00:31.480
tools that are giving AI eyes and hands. And

00:00:31.480 --> 00:00:33.560
we end with a massive hardware breakthrough from

00:00:33.560 --> 00:00:36.079
Cursor. It completely shattered physical speed

00:00:36.079 --> 00:00:38.020
limit. Let's start at the very top of OpenAI.

00:00:38.200 --> 00:00:41.159
We are looking at a deeply complex human situation

00:00:41.159 --> 00:00:43.619
right now. Yeah, there is immense tension. building

00:00:43.619 --> 00:00:45.920
up inside the company. You have explosive commercial

00:00:45.920 --> 00:00:48.859
growth on one side. And a total lack of internal

00:00:48.859 --> 00:00:51.240
trust on the other side. A recent New Yorker

00:00:51.240 --> 00:00:53.579
report really exposes this dynamic. It highlights

00:00:53.579 --> 00:00:55.840
some incredibly serious bad vibes at the executive

00:00:55.840 --> 00:00:58.719
level. Exactly. And the reporting actually draws

00:00:58.719 --> 00:01:02.420
on two unprecedented internal data caches. Right.

00:01:02.539 --> 00:01:04.780
Highly sensitive internal data the public was

00:01:04.780 --> 00:01:07.500
never meant to see. Yeah. So first, there is

00:01:07.500 --> 00:01:11.140
Ilya Sutskaver's 70 -page internal dossier. It

00:01:11.140 --> 00:01:13.659
is packed with internal Slack messages and private

00:01:13.659 --> 00:01:17.579
HR records. Wow. And second, we have Dario Imade's

00:01:17.579 --> 00:01:20.700
highly private safety notes. He kept those during

00:01:20.700 --> 00:01:24.019
his time leading safety there. It feels like

00:01:24.019 --> 00:01:26.480
building a rocket where the engineers fundamentally

00:01:26.480 --> 00:01:29.090
do not trust the pilot. That is a perfect way

00:01:29.090 --> 00:01:31.670
to describe the psychology here. Sup skeevers

00:01:31.670 --> 00:01:35.670
leaked memos make a very specific damaging allegation.

00:01:35.810 --> 00:01:38.310
They claim Sam Ottman consistently misrepresented

00:01:38.310 --> 00:01:41.200
vital safety protocols. Right. He allegedly did

00:01:41.200 --> 00:01:43.560
this to fuel the massive commercial engine. Two

00:01:43.560 --> 00:01:46.079
sec silence. Yeah. That is a profound breach

00:01:46.079 --> 00:01:48.879
of foundational internal trust. The board relies

00:01:48.879 --> 00:01:51.560
entirely on accurate safety data to make decisions.

00:01:51.859 --> 00:01:54.060
It gets even more intense from a broader industry

00:01:54.060 --> 00:01:56.620
perspective. A senior Microsoft executive spoke

00:01:56.620 --> 00:01:59.739
to reporters about this exact dynamic. They compared

00:01:59.739 --> 00:02:02.640
Altman directly to Bernie Madoff or Sam Bankman

00:02:02.640 --> 00:02:04.780
Freed. Wow. Those are historically damaging names

00:02:04.780 --> 00:02:06.620
to be associated with. Right. They actually called

00:02:06.620 --> 00:02:09.120
him a world class scammer. And it completely

00:02:09.120 --> 00:02:11.569
changes. how people view their recent actions.

00:02:11.889 --> 00:02:14.569
Take their new industrial policy for the intelligence

00:02:14.569 --> 00:02:18.090
age. People view this rollout through a much

00:02:18.090 --> 00:02:21.150
more cynical lens now. Exactly. Over a hundred

00:02:21.150 --> 00:02:23.569
inside sources were interviewed for this explosive

00:02:23.569 --> 00:02:26.810
report. They suggest these lofty ideas are a

00:02:26.810 --> 00:02:29.530
classic Altman maneuver. He sets up a structural

00:02:29.530 --> 00:02:33.030
policy on paper first. This effectively deflects

00:02:33.030 --> 00:02:34.889
the current wave of criticism from the public.

00:02:35.050 --> 00:02:37.610
Yeah, but then he quietly dismantles that structure

00:02:37.610 --> 00:02:40.110
once it constrains him. It is a repetitive pattern

00:02:40.110 --> 00:02:43.270
that deeply worries safety experts. It essentially

00:02:43.270 --> 00:02:46.110
paints Sam Altman as the ultimate black box model.

00:02:46.430 --> 00:02:48.870
Right. We simply cannot see how his internal

00:02:48.870 --> 00:02:51.629
decision making actually works. And meanwhile,

00:02:51.810 --> 00:02:53.969
the company marches relentlessly toward a potential

00:02:53.969 --> 00:02:56.650
IPO. They are aggressively eyeing a $1 trillion

00:02:56.650 --> 00:03:01.330
valuation. Yeah. And Elon Musk just updated his

00:03:01.330 --> 00:03:04.289
massive ongoing lawsuit against them. He is seeking

00:03:04.289 --> 00:03:08.069
over $150 billion in legal damages now. He wants

00:03:08.069 --> 00:03:10.169
that money to go directly to the nonprofit arm.

00:03:10.229 --> 00:03:13.110
Exactly. He is also asking a federal judge to

00:03:13.110 --> 00:03:15.569
remove Sam entirely. He wants him forcefully

00:03:15.569 --> 00:03:17.949
kicked off the OpenAI board. What does this mean

00:03:17.949 --> 00:03:20.189
for the public's trust in OpenAI's governance?

00:03:20.650 --> 00:03:23.689
Well, the proposed policies are genuinely great

00:03:23.689 --> 00:03:26.740
for a media headline. But paper structures will

00:03:26.740 --> 00:03:31.039
not fix those deep internal bad vibes. So paper

00:03:31.039 --> 00:03:33.539
structures won't fix deep rooted trust issues.

00:03:33.740 --> 00:03:36.539
Exactly. The human governance must be transparent,

00:03:36.719 --> 00:03:39.580
just like the underlying models. Otherwise, the

00:03:39.580 --> 00:03:42.219
public's deep skepticism is definitely here to

00:03:42.219 --> 00:03:45.479
stay. Let's deliberately shift our focus to the

00:03:45.479 --> 00:03:48.400
broader tech industry now. Despite the messy

00:03:48.400 --> 00:03:50.900
human governance at the top, the technology pushes

00:03:50.900 --> 00:03:54.080
forward. The integration of AI across the industry

00:03:54.080 --> 00:03:57.060
is charging forward ruthlessly. Yeah, the sheer

00:03:57.060 --> 00:03:59.120
adoption numbers we are seeing right now are

00:03:59.120 --> 00:04:02.379
just staggering. Codex just hit 3 million active

00:04:02.379 --> 00:04:04.819
weekly users across the board. That is up from

00:04:04.819 --> 00:04:07.469
2 million users in well under a month. Right.

00:04:07.590 --> 00:04:09.629
The adoption growth curve is basically a vertical

00:04:09.629 --> 00:04:11.930
line at this point. So to celebrate this milestone,

00:04:12.250 --> 00:04:14.050
they're resetting developer rate limits right

00:04:14.050 --> 00:04:16.089
now. And they will systematically reset again

00:04:16.089 --> 00:04:18.170
for every 1 million new users. They plan to keep

00:04:18.170 --> 00:04:20.509
doing this all the way up to 10 million. I still

00:04:20.509 --> 00:04:23.519
wrestle with my own data privacy. Ah, yeah. You

00:04:23.519 --> 00:04:25.839
are definitely not alone in feeling that lingering

00:04:25.839 --> 00:04:28.740
anxiety today. Google just confirmed something

00:04:28.740 --> 00:04:31.899
that deeply validates those exact privacy concerns.

00:04:32.259 --> 00:04:34.579
They openly admitted AI is now crawling your

00:04:34.579 --> 00:04:37.079
most sensitive personal emails. Right. And the

00:04:37.079 --> 00:04:39.100
most crucial detail here is that it happens by

00:04:39.100 --> 00:04:41.540
default. This is a fundamental shift in how personal

00:04:41.540 --> 00:04:44.000
data is handled. If you don't actively intervene,

00:04:44.399 --> 00:04:47.980
a cloud bot reads your entire digital life. It

00:04:47.980 --> 00:04:50.699
is a complete paradigm shift in basic user privacy.

00:04:51.240 --> 00:04:53.800
We are also seeing incredibly deep integration

00:04:53.800 --> 00:04:57.100
across other everyday consumer apps. Like Google

00:04:57.100 --> 00:04:59.839
Maps using Gemini AI quietly behind the scenes.

00:05:00.019 --> 00:05:02.100
Yeah, it automatically writes detailed captions

00:05:02.100 --> 00:05:04.300
for your personal photos and videos. It even

00:05:04.300 --> 00:05:06.819
surfaces your recent photos for easy, frictionless

00:05:06.819 --> 00:05:09.240
posting. Meta is aggressively moving forward

00:05:09.240 --> 00:05:11.540
with their own massive scaling plans. They are

00:05:11.540 --> 00:05:13.939
preparing powerful new models under Chief Officer

00:05:13.939 --> 00:05:16.589
Alexander Wang. This is a truly massive test

00:05:16.589 --> 00:05:19.569
for Meta's long -term strategic vision. It directly

00:05:19.569 --> 00:05:22.850
follows their staggering $14 .3 billion investment

00:05:22.850 --> 00:05:25.550
in stale AI. We also have fascinating new data

00:05:25.550 --> 00:05:29.129
on the underlying hardware powering this. A new

00:05:29.129 --> 00:05:33.350
analytical tool from Epoch maps out global physical

00:05:33.350 --> 00:05:35.889
chip ownership. Right, and the final results

00:05:35.889 --> 00:05:38.110
are exactly what you might expect them to be.

00:05:38.230 --> 00:05:41.790
The dominant answer is literally NVIDIA, NVIDIA,

00:05:41.850 --> 00:05:44.990
NVIDIA across the entire board. Though Google's

00:05:44.990 --> 00:05:49.029
specialized TPUs do appear as very notable challengers.

00:05:49.089 --> 00:05:51.350
Yeah, they definitely do. And the money flowing

00:05:51.350 --> 00:05:53.610
into specialized applications is also staggering.

00:05:54.209 --> 00:05:57.290
Modus just secured $85 million from Lightspeed

00:05:57.290 --> 00:05:59.910
Venture Partners. The rapidly advancing AI -native

00:05:59.910 --> 00:06:01.730
accounting services for traditional business

00:06:01.730 --> 00:06:04.870
sectors. Exactly. It clearly proves AI is hitting

00:06:04.870 --> 00:06:07.970
deeply traditional, strictly rule -based industries

00:06:07.970 --> 00:06:10.850
now. How is a regular user supposed to guard

00:06:10.850 --> 00:06:13.449
their digital footprint amidst this rapid integration?

00:06:14.060 --> 00:06:15.899
I mean, that is the ultimate defining challenge

00:06:15.899 --> 00:06:17.959
for you right now. Default corporate settings

00:06:17.959 --> 00:06:20.019
have flipped from opt -in to opt -out entirely.

00:06:20.379 --> 00:06:22.100
You have to actively manage your privacy settings

00:06:22.100 --> 00:06:24.800
now. Yeah. It is a brand new digital hygiene

00:06:24.800 --> 00:06:27.439
we must all rapidly learn. We are going to take

00:06:27.439 --> 00:06:29.699
a very brief pause here. Insert mid -roll sponsor

00:06:29.699 --> 00:06:32.959
Reed Placeholder here. And we are back to the

00:06:32.959 --> 00:06:36.279
deep dive. We just deeply discussed the massive

00:06:36.279 --> 00:06:39.699
scale of major tech giants. They are constantly

00:06:39.699 --> 00:06:42.259
battling over sprawling server racks and our

00:06:42.259 --> 00:06:44.540
private emails. Right. But a new fascinating

00:06:44.540 --> 00:06:47.660
class of tools is evolving very quickly. These

00:06:47.660 --> 00:06:50.420
powerful new tools give individual users hyper

00:06:50.420 --> 00:06:53.279
specific control and capabilities. We are seeing

00:06:53.279 --> 00:06:56.420
a massive shift toward highly personal digital

00:06:56.420 --> 00:06:58.980
empowerment. Let's carefully unpack some of these

00:06:58.980 --> 00:07:01.019
new empowered individual tools. Yeah. First,

00:07:01.120 --> 00:07:03.939
we have Google AI Edge Eloquent quietly hitting

00:07:03.939 --> 00:07:06.620
the consumer market. It is a completely free,

00:07:06.779 --> 00:07:09.920
entirely offline first audio dictation application.

00:07:10.420 --> 00:07:12.720
It is powered directly by their highly efficient

00:07:12.720 --> 00:07:15.519
localized Gemma models. Exactly. It automatically

00:07:15.519 --> 00:07:17.980
removes your verbal filler words and your awkward

00:07:17.980 --> 00:07:20.139
speaking stumbles. And the fact that it runs

00:07:20.139 --> 00:07:22.660
offline first is huge for data privacy. Then

00:07:22.660 --> 00:07:24.800
we have a truly fascinating system called OpenOwl.

00:07:24.860 --> 00:07:27.120
You give standard AI assistance and... entirely

00:07:27.120 --> 00:07:29.399
new physical abilities. Right. It allows Cod

00:07:29.399 --> 00:07:31.339
and Codex to literally see your live computer

00:07:31.339 --> 00:07:33.579
screen, and it goes way beyond just passively

00:07:33.579 --> 00:07:35.800
looking at the screen, too. It can actively click

00:07:35.800 --> 00:07:38.680
buttons, type text, and navigate across complex

00:07:38.680 --> 00:07:42.040
apps. Yeah. It moves completely seamlessly across

00:07:42.040 --> 00:07:44.459
any web browser you are actively using. We are

00:07:44.459 --> 00:07:47.319
fundamentally moving away from AI as a simple

00:07:47.319 --> 00:07:50.500
conversational chatbot. It is rapidly becoming

00:07:50.500 --> 00:07:53.720
an active proxy operating our physical mice and

00:07:53.720 --> 00:07:56.220
keyboards. Right. And NovaVoice takes that active

00:07:56.220 --> 00:07:58.759
proxy concept even further today. It essentially

00:07:58.759 --> 00:08:01.819
acts as a complete voice OS for your entire machine.

00:08:02.100 --> 00:08:04.560
You can speak naturally to it at over 200 words

00:08:04.560 --> 00:08:07.269
per minute. Which is incredibly fast for natural

00:08:07.269 --> 00:08:09.649
human speech processing. It gives you context

00:08:09.649 --> 00:08:12.649
-aware text generation in absolute real time.

00:08:12.850 --> 00:08:14.769
It remembers absolutely everything and acts directly

00:08:14.769 --> 00:08:17.209
across your entire operating desktop. We also

00:08:17.209 --> 00:08:20.230
have Lessie AI completely changing the modern

00:08:20.230 --> 00:08:22.910
automation landscape. It intelligently discovers

00:08:22.910 --> 00:08:25.189
high -quality professional matches for you across

00:08:25.189 --> 00:08:28.189
the open web. Then it actively automates highly

00:08:28.189 --> 00:08:30.810
personalized outreach and complex follow -up

00:08:30.810 --> 00:08:33.049
messaging. It is really taking over the deeply

00:08:33.049 --> 00:08:34.960
tedious parts of professional... networking.

00:08:35.159 --> 00:08:37.460
Is there a risk in letting AI take the wheel

00:08:37.460 --> 00:08:40.039
on our actual screens? Well, it certainly brings

00:08:40.039 --> 00:08:43.120
massive, unprecedented efficiency to our daily

00:08:43.120 --> 00:08:46.220
workflows, but it inherently requires an immense,

00:08:46.379 --> 00:08:49.340
almost uncomfortable amount of absolute user

00:08:49.340 --> 00:08:52.009
trust. We're delegating the clicking. Not just

00:08:52.009 --> 00:08:54.289
the thinking. Exactly. We are willingly giving

00:08:54.289 --> 00:08:57.490
machines the keys to our physical digital actions.

00:08:57.850 --> 00:09:00.610
Now, all of these amazing new personal tools

00:09:00.610 --> 00:09:03.690
require insane computing power. That directly

00:09:03.690 --> 00:09:06.450
brings us to a massive technical breakthrough

00:09:06.450 --> 00:09:09.190
in physical hardware. It effectively solves the

00:09:09.190 --> 00:09:11.990
biggest structural bottleneck in the entire AI

00:09:11.990 --> 00:09:14.429
industry. Right. We deeply need to talk about

00:09:14.429 --> 00:09:17.409
mixture of experts or MOE architecture. It is

00:09:17.409 --> 00:09:19.830
the undisputed reigning king of modern model

00:09:19.830 --> 00:09:21.950
architecture right now. Let's quickly define

00:09:21.950 --> 00:09:25.129
that dense technical jargon for a moment. Mixture

00:09:25.129 --> 00:09:27.830
of experts means multiple specialized mini models

00:09:27.830 --> 00:09:30.549
working together to solve one prompt. That is

00:09:30.549 --> 00:09:32.570
exactly how it elegantly functions under the

00:09:32.570 --> 00:09:34.850
complex hood. But this specific architecture

00:09:34.850 --> 00:09:37.850
has a legendary, deeply frustrating bottleneck

00:09:37.850 --> 00:09:39.710
holding it back. Do traditional Mowi pipelines

00:09:39.710 --> 00:09:42.009
spend entirely too much precious time just moving

00:09:42.009 --> 00:09:44.590
data? Right. They ironically spend more time

00:09:44.590 --> 00:09:46.690
reshaping data than actually calculating the

00:09:46.690 --> 00:09:50.139
results. It is like stacking Lego blocks of data

00:09:50.139 --> 00:09:53.019
repeatedly over and over. You waste all your

00:09:53.019 --> 00:09:55.019
available energy just moving the heavy blocks

00:09:55.019 --> 00:09:58.299
around. Exactly. So Cursor just proudly introduced

00:09:58.299 --> 00:10:01.100
Warp Decode to fundamentally fix this. It essentially

00:10:01.100 --> 00:10:03.580
eliminates that frustrating structural coordination

00:10:03.580 --> 00:10:06.679
overhead entirely from the active system. How

00:10:06.679 --> 00:10:09.000
does Warp Decode actually accomplish this physical

00:10:09.000 --> 00:10:11.480
feat? Well, instead of traditionally grouping

00:10:11.480 --> 00:10:13.860
work by experts, they completely changed the

00:10:13.860 --> 00:10:16.860
core mapping. They directly map every single

00:10:16.860 --> 00:10:20.759
GPU warp to exactly one specific output value.

00:10:20.919 --> 00:10:23.720
A GPU warp being a basic unit of execution. Right.

00:10:23.820 --> 00:10:26.700
Every single GPU warp computes a single output

00:10:26.700 --> 00:10:29.759
value directly. It intelligently aggregates the

00:10:29.759 --> 00:10:32.279
specific results from all eight routed experts

00:10:32.279 --> 00:10:34.759
seamlessly. And it miraculously does this without

00:10:34.759 --> 00:10:36.840
any intermediate digital staging at all. Yeah.

00:10:37.159 --> 00:10:40.120
They elegantly compressed the entire MoE layer

00:10:40.120 --> 00:10:42.940
down to its bare essence. It is now just two

00:10:42.940 --> 00:10:45.820
heavily fused GPU kernels working perfectly together.

00:10:46.100 --> 00:10:48.220
A fused kernel simply means combining multiple

00:10:48.220 --> 00:10:50.580
mathematical operations into one single step.

00:10:50.720 --> 00:10:52.620
Exactly. It completely removes the traditional

00:10:52.620 --> 00:10:55.100
padding, scatter, and complex combined stages

00:10:55.100 --> 00:10:57.779
entirely. This incredible optimization results

00:10:57.779 --> 00:11:02.279
in 3 .95 terabytes per second of raw memory bandwidth.

00:11:02.750 --> 00:11:05.190
Which is essentially the ultimate physical speed

00:11:05.190 --> 00:11:07.990
limit of the underlying hardware. It absolutely

00:11:07.990 --> 00:11:10.909
maximizes the sheer potential of the Blackwell

00:11:10.909 --> 00:11:13.610
HBM3 hardware architecture. Two sec silence.

00:11:13.870 --> 00:11:17.070
Oh, imagine maxing out the physical speed limit

00:11:17.070 --> 00:11:19.169
of the hardware. It is an absolutely incredible,

00:11:19.490 --> 00:11:21.789
almost unbelievable feat of software optimization

00:11:21.789 --> 00:11:25.549
meeting physics. But the pure speed is only half

00:11:25.549 --> 00:11:28.070
of this amazing technical breakthrough. The final

00:11:28.070 --> 00:11:30.809
output is actually 1 .4 times closer to the full

00:11:30.809 --> 00:11:33.549
mathematical reference. Right. It rapidly approaches

00:11:33.549 --> 00:11:36.509
a full FP32 reference compared to standard everyday

00:11:36.509 --> 00:11:39.230
decoder. Why is getting both speed and accuracy

00:11:39.230 --> 00:11:42.169
such a unicorn event in AI? Because usually...

00:11:42.399 --> 00:11:44.860
In the relentless race for speed, we begrudgingly

00:11:44.860 --> 00:11:47.639
accept a quality tax. Unprecedented efficiency

00:11:47.639 --> 00:11:49.960
typically demands severe data compression from

00:11:49.960 --> 00:11:51.980
the software engineers. We relentlessly shrink

00:11:51.980 --> 00:11:54.299
models down to 4 -bit or 8 -bit quantization

00:11:54.299 --> 00:11:57.340
formats. Yeah. We quietly sacrifice that nuanced

00:11:57.340 --> 00:12:00.019
depth just to get extra tokens per second. Usually

00:12:00.019 --> 00:12:02.700
more speed demands a sacrifice in accuracy. But

00:12:02.700 --> 00:12:05.679
wonderfully, not this time. Cursor's warp decode

00:12:05.679 --> 00:12:08.980
is a genuinely rare, highly impactful scientific

00:12:08.980 --> 00:12:11.679
breakthrough. It tangibly improves mathematical

00:12:11.679 --> 00:12:14.159
accuracy while nearly doubling the operational

00:12:14.159 --> 00:12:17.580
processing speed. Right. It is a massive, unqualified

00:12:17.580 --> 00:12:20.759
structural win for the future of scaling. Let's

00:12:20.759 --> 00:12:23.200
slowly zoom out and look at the truly big picture

00:12:23.200 --> 00:12:26.299
here. We are actively living in a deeply fascinating,

00:12:26.580 --> 00:12:29.320
highly complex modern paradox. We really are.

00:12:29.480 --> 00:12:31.980
On one side, we have these absolutely flawless,

00:12:32.240 --> 00:12:35.360
highly optimized digital machines. They quietly

00:12:35.360 --> 00:12:38.720
operate at 3 .95 terabytes per second of pure,

00:12:38.879 --> 00:12:41.519
unadulterated mathematical logic. Right, without

00:12:41.519 --> 00:12:44.000
any intermediate staging or clumsy coordination

00:12:44.000 --> 00:12:46.919
overhead at all. And on the other side, we have

00:12:46.919 --> 00:12:49.080
deeply fundamentally flawed human beings. We

00:12:49.080 --> 00:12:51.940
have the messy, bad vibes corporate governance

00:12:51.940 --> 00:12:54.039
actively building these... digital machines.

00:12:54.200 --> 00:12:55.960
It really makes you pause and wonder about our

00:12:55.960 --> 00:12:58.360
own internal human structures. It deliberately

00:12:58.360 --> 00:13:00.519
leaves us with a deeply provocative thought for

00:13:00.519 --> 00:13:03.639
you today. If a machine can be perfectly optimized

00:13:03.639 --> 00:13:07.220
to run without coordination overhead, why is

00:13:07.220 --> 00:13:09.320
human corporate governance so inherently prone

00:13:09.320 --> 00:13:12.179
to bottlenecks and black boxes? That is a profound

00:13:12.179 --> 00:13:15.200
question. We ask you to deeply consider what

00:13:15.200 --> 00:13:18.139
true structural transparency actually looks like,

00:13:18.259 --> 00:13:20.860
not just in the elegant code quietly running

00:13:20.860 --> 00:13:23.980
on our fast servers, but crucially in the closed

00:13:23.980 --> 00:13:27.320
boardrooms that actively control our shared digital

00:13:27.320 --> 00:13:29.919
future? Yeah, that is a vital question we all

00:13:29.919 --> 00:13:31.960
desperately need to start asking. Thank you for

00:13:31.960 --> 00:13:34.019
actively joining us on this deep dive today.

00:13:34.379 --> 00:13:36.960
Keep rigorously questioning the surrounding systems,

00:13:37.120 --> 00:13:39.980
both the cold silicon and the messy human. We

00:13:39.980 --> 00:13:41.240
will thoughtfully see you next time.