Latest / Elon Musk Podcast / Elon Musk Neuralink Update

Transcript

Elon Musk Neuralink Update · Elon Musk Podcast

0:03
Welcome, I hope this is working. Welcome to neural link up live
0:10
update. We're going to tell you about
0:13
the progress of the first patient with neural link and
0:19
sort of your recap of of the progress there.
0:23
Then talk about what changes we were making for the second
0:27
patient, which we're hoping to do an implant in the next week
0:31
or so. And this is for our first
0:33
product, which is called Telepathy, which enables you to
0:36
control a computer or a phone just by thinking.
0:39
So let's in fact. So we'll start off with just
0:42
some introductions. DJ you want to start?
0:44
Hi everyone, My name is DJ Saul. I'm an electrical engineer and a
0:47
chip designer. By training.
0:49
I led the design of first several generations of the
0:54
Neurolink implant. Currently I was on the founding
0:57
team and currently a President. I'm Matthew McDougal.
1:00
I'm a practicing neurosurgeon and head of Neurosurgery at
1:03
Neurolink. Yeah, go ahead.
1:06
Yeah. Head of Branding Interfaces
1:09
applications. And I'm bliss.
1:11
I'm a software engineer at Neurolink trying to figure out
1:12
how to turn brand activity into cool stuff in the world.
1:16
All right. Thank you.
1:16
Well, let's see. So we'll just get going into the
1:19
presentation. So our first product is sort of
1:25
like I said, we call telepathy, which is enables the person with
1:30
Neuralink implant to control their phone or computer just by
1:33
thinking. And once you can control your
1:35
phone or computer, you can essentially control almost
1:37
anything just and literally just by thinking.
1:39
So there's no eye tracking or anything, it is purely, purely
1:45
your thoughts. So this is really quite a
1:50
profound device that can help a lot of people who have lost the
1:54
connection between their brain and body.
1:56
So imagine people like Stephen Hawking, who, you know, imagine
2:03
if he if he could communicate at the same speed as someone who
2:06
had still had the connection to their brain and body.
2:08
So it's really something that can help millions of people
2:12
around the world. And it's a, it's part of our
2:16
overall goal of enabling a very high bandwidth connection
2:21
between the brain and, and your and the rest of the world and
2:25
your computers. The long term goal to which
2:28
sounds a little esoteric, is to mitigate the the risk of, of the
2:33
civil civilizational risk of AI by having a sort of closer
2:40
biosis between human intelligence and digital
2:43
intelligence. But that, that'll take many
2:46
years along the way, we're, we're going to help solve a lot
2:50
of brain injury or spinal injury issues.
2:55
So, and then with our first product apathy, that's, that's
3:00
going to be really quite profound that there is also
3:04
potential long term for bridging the gaps or if there are damaged
3:09
or severed neurons being able to span the gap between the brain's
3:14
motor cortex to the spine to enable someone to use their body
3:18
again. I think that would be very
3:20
exciting. And it's, you know, that that is
3:24
something that is possible in the long term.
3:29
And then our second product, which we've demonstrated to work
3:34
with monkeys is Blind side, which would enable someone who
3:38
is completely blind or lost both eyes or completely lost their
3:42
optic nerve to be able to see. So that's that's something that
3:48
we hope to demonstrate in the future.
3:55
So let's just give you a sense of what the device is a way to
3:59
think about the neural link device is kind of like a a
4:01
Fitbit or an Apple Watch with with tiny wires or electrodes.
4:06
Those those tiny wires are implanted in the in the brain
4:10
and they read and write electrical signals.
4:14
So a lot of people think the brain is incredibly mysterious
4:19
thing. It's it, it is mysterious in a
4:21
lot of ways, but but it is actually, it does operate with
4:24
like electrical signals. So if you can read and write
4:26
those electrical signals, you can interface with the brain.
4:29
And the devices is sized so that it is the same size as the as
4:37
the piece of skull that is removed.
4:38
So it's like a few centimeters diameter of skull that's
4:43
removed. We replaced that with the device
4:46
after implanting the tiny wires with a surgical robot and that
4:50
enables read write capability to the neurons.
4:54
Completely wirelessly. Yeah, yes, exactly.
4:57
It's completely wirelessly. So like I could I could have a
5:01
neural link right now, you wouldn't know.
5:04
And it charges inductively. So you could just basically have
5:08
an electromagnetic ad that you charge the device with.
5:13
So yeah, it's like an Apple Watch.
5:15
Exactly so. Except that it's actually a much
5:19
harder technical challenge to solve given that there's limit
5:23
as to how much heat the brain tissue whereas in for phones.
5:26
And have a hot watch. Or whatever you really care how
5:28
much if it's sitting on a table. Sure.
5:31
Yeah, it's also it's got to go through scan and stuff as well
5:33
in our case. So it is a tougher challenge to
5:36
to charge and to have I bandwidth communications given
5:40
that it's got to go through skin and hair and stuff.
5:43
We have solved it. We have solved it, yeah.
5:48
So yeah. Yeah.
5:49
So our first step with the telepathy is basically to unlock
5:54
digital independence for people with policies and to allow them
5:59
to control the computer just with their mind without moving
6:02
the body. And our goal is to provide them
6:05
the same level of control, functionality and reliability
6:09
that I have when I'm using a computer, even better than the
6:12
level of control I have. And it's not a high bar for.
6:15
You just to be clear, this guy, he's controlling this with his
6:19
brain, so he's not like you can't see his hands in this
6:21
video, but he's not using a mouse and keyboard.
6:23
Just, you know, thinking about how to move the cursor and
6:26
playing Civilization. No eye tracker.
6:28
Right, there's no eye tracking from I.
6:29
Mean he's lost. He's lost Junior.
6:31
He's like, he's going to watch this on Twitter.
6:32
Just thinking that's it. Just.
6:34
Thinking like just a couple days.
6:35
Like cursor move here. Yeah, yeah, yeah.
6:37
This is like last night or two nights ago something.
6:39
Yeah. I think I think the way he also
6:41
described it is he's using the. Yeah.
6:44
He has many more videos on his on the platform.
6:48
Definitely check them out. Yeah, so he can, he's streaming
6:51
that live and also can talk and like move his head without a
6:54
problem multi time. Yeah, you guys like if you join
6:57
this live stream, you can ask him questions, he'll he'll tell
6:59
you all about what it's like to move.
7:02
Also I think I haven't played civilization myself but I think
7:06
this is actually not easy mode. This is expert this.
7:08
Is emperor mode emperor mode? If you have played stiff emperor
7:11
mode, it's like the highest difficulty level.
7:12
Just the point is like this is a commonly demanding task while
7:15
live streaming playing the hardest mode of team and he's
7:18
able to do that while moving. Talking, engaging with you know
7:22
the audience while playing. One of the other games he likes
7:25
to play a lot is chess. I think it gets lost sometimes
7:28
that he's actually playing speed chess.
7:30
Against me, yeah. Which requires an incredibly
7:33
high fidelity degree of control and and speed of control in
7:38
order to be able to win. So also another cool stuff about
7:45
about our device is that we can use it anywhere, anytime, also
7:49
on a plane, during a flight, while creating really cool means
7:53
of care. Also, our device unlocks things
7:57
that previously were impossible for our participants.
8:01
For example, we're able to connect him to his gaming
8:05
consoles, which they Mario Kart with friends and family, and it
8:08
was lovely to see them playing together after years that he
8:12
couldn't do it since he's injured.
8:15
Imagine if you're sitting umbrella over from the sky on a
8:18
plane. You look over, he's making a cat
8:20
meme. No hands, no movement.
8:23
Yeah. Live in a real world.
8:26
Yeah, it's strange, strange time.
8:29
Yeah. And he loves using the device
8:32
and using independently daily to watch videos, read, play games
8:38
using the browser. And the key metrics that we care
8:42
is to make sure our device is actually useful is to is
8:45
basically the amount of hours we use the device daily and weekly.
8:49
And we track it weekly since the since the surgery and on weeks
8:53
that he's not too busy and not travelling, he can even reach 70
8:57
hours of using the device a week.
9:00
This is amazing and he would of course love to use it more, but
9:04
need to run recent sessions, he needs to sleep sometimes and
9:09
also of course to charge the device once in a while.
9:11
Hopefully we'll improve that over time, I think.
9:14
That maybe not obvious to people who are watching this.
9:16
Like it's a normal MacBook he's controlling.
9:17
This isn't like some limited efficient thing, but there's
9:20
only a few options like he can just do anything that you can do
9:22
on a MacBook Pro. Same one I have on my desk.
9:24
Actually, it's the exact same one.
9:29
And maybe another interesting point is that on the first day
9:33
he used BCI brain control, he was able to break the previous
9:38
world record for cursor control is by using the brain.
9:42
And recently he even doubled it and was able to outperform about
9:47
10% of our engineered neurolink. And you can be sure that we are
9:51
very good in this game and very quick.
9:53
And if you want to check out how well, how well you can do it,
9:56
you can do it on our website. And it's very addictive games.
10:01
Yeah, it's a very simple game. You just have to click on the
10:02
square. But but it's it's it's it's
10:07
actually, even though it sounds silly, it's it's quite a yeah,
10:11
it can be quite a it sounds like it could be quite addictive.
10:13
And it's especially if you get a low score and you think there's
10:16
no way I got a. So I, I mean, anyone who wants
10:19
to try this, I recommend going to theneurolink.com website and
10:23
seeing, seeing if you can beat Nolan's record.
10:25
And it's that you will find that's actually quite difficult
10:27
to do so. And this is really with version
10:31
one of the device and with only a small percentage of the
10:34
electrodes that are that are working.
10:36
So this is, this is really just the beginning, but even the
10:40
beginning is twice as good as the world record.
10:43
This is important to emphasize the, you know, the media has a
10:48
habit of of saying that the glass is 10% empty, but but
10:53
actually it's 90% full. So I think it's really quite an
10:57
accomplishment of the neural link team to have achieved with
11:00
first with the first patient, the first device twice the world
11:05
record for the range computer bandwidth.
11:09
That's really an astonishing, an amazingly great outcome and it's
11:16
only going to be get better from here.
11:17
So the potential is to ultimately get, I think to
11:21
megabit level. So that's, that's part of the
11:26
long term goal of, of improving the, the bandwidth of the brain
11:29
computer interface. If you think about like how low
11:33
the bandwidth normally is between a human and a device,
11:37
it's the average bandwidth is extremely low.
11:38
It's it's, I said less than one bit per second over the course
11:43
of a day. If there are 86,400 seconds in a
11:45
day, you're outputting less than that number of bits to any any
11:50
given device except in perhaps very rare circumstances.
11:54
So the this is actually quite important for for AI, you know,
12:05
basically for for human AI, symbiosis is just being able to
12:11
communicate at A at a speed. Yeah, I can follow.
12:19
So yeah, just to emphasize again, he's performing at this
12:23
extremely high level with about 15% of his channels functional.
12:30
And so we want to mitigate any of the problems that led to that
12:33
situation. So, you know, the brain is a
12:37
fascinating organ. Share with you some of the
12:39
secrets about the brain. During any typical brain
12:42
surgery, a small amount of air is introduced into the skull.
12:47
That's because neurosurgeons like to have as much room as
12:50
possible around the brain. And so there's this little known
12:54
control mechanism of allowing the CO2 concentration in the
13:01
blood to rise a bit, which allows the brain to either
13:06
expand or contract depending on where you target that CO2.
13:10
But typically neurosurgeons will have the brain shrink by
13:13
lowering CO2. What we're going to do in our
13:16
future surgeries is keep the CO2 concentration actually quite
13:20
normal, maybe even slightly elevated, and that'll allow the
13:23
brain to stay its normal size and shape during surgery.
13:27
That should eliminate this air pocket that we saw in the first
13:29
participant. Now that air pocket we think may
13:32
have contributed to eating up some of the thread slack is as
13:37
the air bubble migrated to be under the implant, push the
13:40
brain away from the implant. And so that's easy enough to
13:44
fix. Another consideration that we
13:48
want to focus on for our upcoming participants is that
13:54
the brain, think of it like a really complex folded onion.
13:57
It's layer upon layer of sheets of neurons all over the surface
14:02
of the brain folded into this, you know, odd looking shape.
14:08
The folds of the brain travel down deep into the brain and and
14:12
along with it go those onion layers of neurons.
14:16
And if we insert very close to one of the folds where there may
14:20
be very useful information encoded in neurons, we may end
14:24
up travelling with our threads parallel to some of the layers
14:28
of neurons that we're most interested in, avoiding them
14:30
entirely. To avoid that possibility, we're
14:33
going to insert in our future participants more close to the
14:37
middle of the apex of the folds, ensuring that we're crossing the
14:43
layers of interest, layer five of the cortex.
14:45
I also think that it's important to highlight here those tiny
14:50
wires that Elon mentioned, They're fraction of a human
14:53
hair. They're very flexible,
14:55
intentionally so, because brain is constantly moving and you
14:58
want the electrodes to be moving with the brain, causing less of
15:02
the scarring. And it's actually impossible for
15:07
a human neurosurgeon, however talented Matthew is to actually
15:10
maneuver them by. So we have a surgical robot that
15:13
we built that can actually precisely target them in any 3
15:16
dimensional space, XY as well as Z with Micron level precision
15:20
while avoiding vasculature so that you don't disrupt the the
15:23
and and cause immune response from happening.
15:25
So we we actually have the technology to be able to place
15:28
them exactly where we want them in three.
15:31
Yeah, it was truly amazing to see the surface of the brain
15:34
after the robot had inserted all the electrodes on the 1st
15:37
participant without a drop of blood.
15:39
Insight. It's really quite an
15:42
achievement. Yes, something that probably
15:45
most people don't realize is that the the brain appears to be
15:48
sort of somewhat undifferentiated.
15:50
So if you look at the cortex, it looks like a whole bunch of
15:53
folds that where, you know, maybe like it's, it's not
15:58
obvious, just looking at a say, a picture of the brain that,
16:01
that, that it's the brain is highly differentiated.
16:03
That there's you pretty much know exactly where the part of
16:08
the brain is that controls your right hand and your left hand
16:12
and your leg. And like that kind of thing or
16:15
vision, it's actually quite precisely located.
16:19
It's not some people like might, might think, look at the brain
16:23
like, oh, it could be, it could be anywhere.
16:24
But actually we it's your brain is highly differentiated, even
16:30
though it doesn't look it's yeah.
16:34
Do you want to describe how we actually where like how we
16:37
identify where to drill the? Yeah.
16:38
So we can we can put a patient that is considering this implant
16:44
into an fMRI, so a functional magnetic resonance imaging
16:47
machine and ask them to imagine hand movements that, you know,
16:51
because of the spinal cord injury don't happen.
16:54
But just imagining those hand movements causes these areas of
16:57
the brain to light up in the fMRI scanner.
17:00
And so we have a pretty good idea based in, in fact, for each
17:05
individual participant, which part of their brain is going to,
17:10
you know, respond to imagined movements of the hand.
17:14
And so we can map those imagined movements, much as we all do
17:17
when moving a mouse to controlling a cursor on a
17:20
screen, even without the use of a mouse.
17:25
Yeah. But anyway, I think it's kind of
17:26
an important point that like, it's not like you're the part of
17:29
your brain that controls your hand might be anywhere in the
17:31
cortex. It's this is not the case.
17:33
It's going to be in a very specific region and it's going
17:35
to be extremely common across people.
17:38
Precision is key too. Yeah, the left-handed,
17:43
right-handed thing in my mind too.
17:45
Like if you're right-handed, you want the device on the left
17:48
side, the actual lateral side to the hand.
17:50
That's atomic. Yeah, the left side of your
17:52
brain controls right side of your body.
17:53
Everything's crossed. Another of the risk mitigations
17:57
we're looking at in the future is that, you know that the
18:01
implant has a certain size, the depth of the bottom of the
18:04
implant actually thinner than the average human skull.
18:08
And So what we want to be able to do is control the size of the
18:13
gap under the implant. If the threads that travel from
18:18
the implant into the brain as much slack as possible.
18:22
We didn't do this in the first participant because we didn't
18:26
want to, you know, manipulate any of their tissue that we
18:29
didn't absolutely have to in upcoming implants.
18:32
Our plan is to or sculpt the surface of the skull very
18:37
intentionally to minimize the gap under the implant such that
18:41
the bottom of the implant travels perfectly flush with the
18:45
normal contour of the inner side of the skull.
18:48
That will put the implant closer to the brain, eliminate some of
18:52
the tension on the threads and we think it will reduce some of
18:56
the tendency of threads to retract, right and.
18:59
We actually built a tool to do right.
19:01
Yeah, this is, this is actually, this is a very important detail.
19:05
You really want the the inner contour of the skull to be flush
19:11
so that the implant doesn't there's no the brain doesn't
19:14
want to pucker up into the into the gap.
19:16
That's really quite a big deal. So that like like minimize the
19:22
air pocket and the implant being flush with the the inside
19:27
contour of the skull is are two very important improvements.
19:31
The additional benefit here is that you know you do see some
19:35
amount of stick up, what we call stick up, so you minor bump in
19:38
the head, but this actually eliminates it even further.
19:41
Yeah, yeah. I mean, it's like really our
19:43
goal is that, that if you run your hand over the top of the
19:48
skull, you don't feel any, any bump, you don't feel any, any
19:50
device. And that even if someone was
19:53
bald, you wouldn't really even notice it.
19:55
And and then from the the inner inner contour of the skull that
19:59
the the brain or physical standpoint doesn't really notice
20:02
that there's a divot in the skull because there's no divot.
20:14
OK. Another aspect of the of the
20:17
human brain that you know, obviously differs from any of
20:20
the animals that we tested in is that the human brain is a lot
20:23
bigger. And so you may not realize that
20:26
that means that the human brain moves quite a bit more than any
20:32
of these other smaller brained creatures.
20:34
And so when we open the skull, we see the brain travel toward
20:40
and away from the robot about 3mm in total as the heart beats
20:43
and and the breathing takes place.
20:46
And so that movement, you know it, it adds a small challenge
20:50
for the robot in precisely choosing a depth to insert each
20:54
thread. It's not an enormous challenge.
20:57
And we've already upgraded the robot's capabilities to be able
21:00
to even more precisely target depth in, in even a very rapidly
21:06
moving brain with a high amplitude of movement.
21:13
You may think the most obvious mitigation for threads that
21:15
pulled out of the brain is to insert them deeper.
21:17
We think so too. And so we're going to broaden
21:22
the range of depths at which we insert threads.
21:25
So, you know, for the very first participant, we had an enormous
21:29
amount of data from our animal work and we had very highly
21:33
optimized our insertion depth to maximize the crossing of the
21:39
layers of interest in the cortex with the electrodes that we're
21:42
recording from. Now that we know retraction is a
21:46
possibility, we're going to insert at a variety of depths
21:51
that even in several cases of differing amounts of retracting
21:55
threads, we're going to have electrodes at the proper depth
21:59
and with the deepest threads be able to track how much
22:03
retraction has occurred across the surface of the brain from
22:07
from each thread. And so we're going to, you know,
22:09
both have more threads in the right layer and have better data
22:13
on how much retraction has occurred.
22:15
If you're a BCI nerd, you might know that being able to control
22:18
individual Z depth per thread is not something that most neural
22:23
interface devices offer. Most neural interface devices
22:25
are kind of a static fixed rigid array that you push in and all
22:28
the electrodes are on depth. Right.
22:30
To be able to do this is actually pretty pretty novel.
22:32
Part of the robot. Yeah, the historical approach is
22:36
to actually pound in a sort of bed of nails with an air hammer
22:39
into the. Brain, it looks crazy that that
22:41
that is, Yeah, just with a with a pneumatic hammer.
22:47
That's the that's the this is it sounds come somewhat barbaric.
22:49
This is not what we do, but this is the what's been done before
22:52
is. They're literally just hammering
22:55
in what looks like a bed of nails with the brain.
22:57
Which? Actually works.
22:58
It's astonishing that it actually.
22:59
Works, but I mean some people like manual like DBS probes.
23:02
You're just sticking in by hand. Our research is just guiding
23:05
them in. Those are several, several
23:07
orders of magnitude more volume of brain tissue that you're
23:10
destroying compared to what we're doing.
23:12
But that deep brain simulation stuff does actually work, and it
23:14
actually helps people a lot. Yeah, yeah, yeah.
23:18
That's a great product. Yeah.
23:20
But I mean, I think we'll we'll be able to do a much more
23:23
finessed version of that down the road.
23:26
So I mean, it's really difficult.
23:30
Like the the neural link device is something that really
23:32
absolutely minimizes damage to the brain, absolutely minimizes
23:36
the load on the patient. And the goal is to allow someone
23:40
to live a completely normal life.
23:42
They work that you won't even notice that someone even has the
23:45
device. So like I said, we're restoring
23:49
the ability to control your computer and phone.
23:51
That's how telepathy and then next device being able to allow
23:56
people to see that cannot see before.
23:58
And in fact, you could allow people to see kind of like Dora
24:00
the Forge in Star Trek in any but whatever.
24:03
Infrared. Yeah, infrared, ultraviolet,
24:07
radar. So.
24:12
So I think another way of saying it is that we want to give
24:16
people superpowers. So it's not just that we're
24:19
restoring your prior brain functionality, but that you
24:22
actually have functionality far greater than a normal human.
24:26
That's a super big deal. And I also think, you know,
24:29
often times the questions that we get a lot is why do you have
24:31
to actually go into the brain? What if you place it on the
24:34
surface or outside the skull? Basically the Long story short,
24:38
the physics of how it works, you really need to get the sensors,
24:41
which are these facing in the brain next to the source which
24:47
you're on as close to it as possible.
24:49
Otherwise what you get is you get a population response and
24:52
not be able to kind of do the level of controls that we
24:55
believe of. Yeah, I mean, a good sort of
24:59
analogy would be like if you're trying to understand what goes
25:02
on in a factory, you kind of need to go into the factory.
25:04
You can't just put a stethoscope on the wall and try to figure
25:07
out what's going. Like anything on the outside of
25:09
the trying to read things from the outside is like putting a
25:13
stethoscope on the wall of a factory, trying to understand
25:15
what's going in the factory. It's not going to be effective.
25:19
You've got to be threads. You got to be in there.
25:23
So. But I just want to be emphasized
25:26
again, like the goal is to give people superpowers, not just to
25:31
restore prior functionality. So it's very exciting.
25:35
And I think that should give hope to a lot of people in the
25:39
world that the future is going to be exciting and inspiring and
25:43
the technology is going to give them superpowers.
25:46
I mean, that's that's amazing. Yeah, I guess.
25:59
These. Off yeah and could can you
26:01
multitask but yeah in fact if you look at Nolan's streaming
26:04
and you can just check out Nolan's streams on on the X
26:08
platform he's multitasking all the time so he's playing video
26:12
games while talking and. Listening to podcasts?
26:15
Nice listening to podcasting. Yeah.
26:18
Yeah, exactly. So it's really just like if you,
26:21
you're using your hands and you, you can be, you know, playing a
26:24
video game while talking so. I mean, don't take that word for
26:27
it. Just go watch.
26:27
I mean, yeah, yeah, he's out there on the Internet doing his
26:29
thing. Yeah, yeah, exactly.
26:33
So can he do keyboard shortcuts or is it just a mouse?
26:35
Yeah. That that's actually what we're
26:36
working on right now. Sure, so currently he's working
26:40
with the mouse, but we are also exploring and decoding more
26:43
dimensions from the neural activity.
26:45
Multiple clicks. So to do shortcuts or just able
26:50
to control more games, control games with an Xbox controller.
26:57
But also in the future we expand, we plan to expand to
26:59
decode text, not just the mouse control, but also allow our
27:03
participant to type much faster. And yeah.
27:09
Yeah, actually. So maybe going back to the
27:11
discussion of thread retraction, you know, one of the very
27:13
exciting parts to me about this story is that we're able to do
27:15
so much with 15% of channels. You have more channels.
27:18
What that actually offers you is not just faster mouse control,
27:21
because in the motor cortex, neurons don't all represent the
27:23
same thing. So if you're trying to
27:25
understand, like, you know, what an individual finger is trying
27:28
to do, you might or might not have an Electro next to it.
27:31
And the more channels you have in the brain, the higher
27:32
likelihood you have, you know, representation or decodability
27:35
of all fingers on the hand. It's like you're trying to do
27:38
something like output text at a fast rate.
27:40
It's something that matters a lot for people who are
27:41
completely locked in, who cannot speak at all, who are trying to,
27:44
you know, just say I love you to them, to a loved one in their
27:46
family or ask for a glass of water or a scratch or whatever.
27:49
You know, being able to type at a faster rate, it's extremely
27:51
important. And the more fingers you have
27:52
access to, higher probability, you can do that efficiently.
27:56
And so, yeah, you know, I'm super excited about how high the
27:59
ceiling is we can that we can get to as we resolve this
28:02
standard traction issue. Yeah, I mean we're like we're
28:05
currently at approximately 1010 bits per second be great.
28:08
But ultimately we want to get to megabit and I think say
28:13
ultimately whole brain interface.
28:15
I think you know, many years from now I think Gigabit level
28:19
is possible. So that's that's pretty
28:23
astonishing. Now you know with this is still
28:27
version one of our device, as we mentioned, it's version one with
28:30
only 15% of the threads working. The the current device has 64
28:36
threads with 16 electrodes on each thread.
28:39
Our next device has 128 threads with, with eight electrodes per
28:44
per thread. Because as we get more confident
28:47
about how, where exactly to place the the electrode the the
28:52
thread, you need fewer electrodes per thread.
28:55
So we can essentially with the current device without
28:57
substantial changes potentially double the bandwidth if we
29:03
accurate with the with the placement of of the threads and
29:06
then our next generation device will have maybe even.
29:09
More channels, Yeah, yeah. So yeah, so next device for
29:13
every four, yeah, 3000 channels. So this will just keep getting
29:18
better and better really moving up I think in orders of
29:21
magnitude in factors of 10 basically in in what kind of
29:25
bandwidth. So I think it won't be, it won't
29:29
be all that long before someone with a neural link device can
29:33
communicate faster than someone who is has a fully functional
29:36
body. And yeah, so I think, you know,
29:41
faster than the fastest speed typist or auctioneer.
29:45
The E sports tournaments are. Going to be won't be able to
29:48
speak faster than someone can communicate with a neural link
29:52
telepathy device. It may be a very interesting
29:54
part of this. Basically, we currently connect
29:58
standard inputs to the computer through mouse and keyboards, but
30:02
very soon as we will have a much part of bandwidth, we need to
30:05
think about new ways to actually build the interface for the
30:09
devices. This is something that we can't
30:13
accept. Yeah, no, that's, that's a good
30:15
point because the the current import devices are centered
30:20
around human hands. Yeah.
30:21
So it's like we've got these, you know, little meat sticks
30:24
that we move and the this certain rate at which you can
30:29
move your little move your fingers.
30:31
And, and so we've got like the mouse and the keyboard and with
30:33
a joystick control, you know, like Xbox controller or
30:36
something like that. But you really don't need that.
30:39
You can actually, you don't, you don't need, since you're no
30:43
longer, if you're not trying to use your hands, you don't, you
30:46
actually don't need those conventional control mechanisms.
30:53
And so This is why like, ultimately I think you'll be
30:55
able to do conceptual telepathy like where you can communicate
31:01
entire concepts uncompressed to someone else with a neural link
31:06
or to the computer. Even today we have some problems
31:09
here where like, you know, if you don't feel the mouse
31:11
clicking under your finger, how do you know it actually
31:13
happened? Because you know, you're, you're
31:14
seeing it on the screen, but you don't actually feel the mouse
31:16
clip. You don't have the
31:17
proprioceptive feedback of, you know, the keys under your
31:19
fingertips or the track pad under your.
31:21
There's all sorts of interesting UX challenges, how to actually
31:24
give the user some sense of what their decoder's actually doing,
31:27
or what the Neurlink's actually doing.
31:29
I mean, they're trying to use. So wireless.
31:38
Yeah, it's Bluetooth. Just a Bluetooth connection,
31:41
just like how your normal Apple mouse or like Apple Magic
31:44
Keyboard connects to your computer.
31:45
Same exact thing. In fact in Yeah, we can
31:49
basically have this exposed as an HID interface if we want.
31:52
HID is just the name of the protocol for like sending fits
31:54
my mouse into a computer. Yeah, I can plug into basically
31:58
anything. Yeah.
32:00
I mean, I think we, we chose that interface because it's
32:02
ubiquitous. Yeah, basically any devices are,
32:05
are have Bluetooth capabilities. Our our long term goal is to
32:09
actually have our own protocol, you know, that is safe and
32:11
secure. But for now, you know we've
32:13
chosen it for interoperability. So the question is, can a neural
32:18
link chip repair the paralysis in the long term?
32:21
You know, we can't do that right now.
32:23
We have done sort of preliminary work implanting a second neural
32:27
link in the spinal cord and we can restore naturalistic looking
32:34
and, and leg movements in animal models.
32:39
But this isn't something that is, you know, don't, don't hold
32:42
your breath waiting for it. It's going to be a while.
32:43
We've got a lot of work to do. But yes, there's no reason in
32:48
theory that we can't repair paralysis.
32:51
Yeah. I mean essentially to, to, I
32:56
mean, it there's, there's no, there's no physics barrier to
33:00
fully solving paralysis. That is perhaps a way to say it
33:03
that you've got signals coming from your motor cortex that if
33:09
they are transferred past the point where the the nerves are
33:14
damaged, essentially just it's basically a communications
33:17
bridge. So you bridge the communications
33:19
from the motor cortex past the the point in the neck or spine
33:24
where the nose is damaged and you should like it is physic it
33:28
is possible from a physics standpoint to restore full body
33:31
functionality from a physics standpoint, it's a very hard
33:34
technical problem, but it but it there is nothing that prevents
33:38
it happening from a physics standpoint.
33:45
So in terms of next phase of the rollout, well, we really want to
33:49
make sure that we make as much progress as possible between
33:52
each neural link patient. So this is we're only just
33:55
moving now to our second neural link patient, but we we hope to
34:00
have, you know, if things go well high single digits this
34:04
year. And I don't know maybe if this
34:09
is somewhat dependent on regulatory approval and how much
34:12
technical progress we make, but within a few years, hopefully
34:16
thousands. Yeah.
34:18
And I think one thing that is important to highlight is that,
34:22
you know, it's not that we've built only one device and one
34:24
surgery. We've done hundreds of surgery.
34:26
We've built thousands and thousands of devices even for
34:30
just the the ability to unearth any sort of low frequency
34:33
failure mode. So we have already been
34:36
investing very heavily in infrastructure to be able to
34:39
scale this thing on the device manufacturing side as well as on
34:42
the surgery side of things. We want to be able to help as
34:44
many people as quickly as well, yeah, go through obviously the
34:47
appropriate hurdles that are regulatory challenges and
34:51
proving out the device with. Yeah, and the, the, the device
34:56
implantation really needs to become almost entirely if not
34:59
entirely automatic in the same way that, say, LASIK eye surgery
35:03
is done. You know, you don't have an
35:06
ophthalmologist with a, a laser cutter by hand that that would
35:09
be crazy. But the ophthalmologist overseas
35:13
the, the LASIK machine and make sure that the settings are
35:16
correct and then the machine does everything and restores
35:19
your eyesight. It's really remarkable how how
35:22
many people have had their eyesight restored with, with
35:25
LASIK. And I think there's another one
35:26
called Smile. It's they, they keep making it
35:28
better. We need to have something
35:30
similar for a Neuralink implantation so that you
35:33
basically sit down and whatever the, the, whatever, whatever
35:36
kind of upgrades or, you know, brain fixes are needed.
35:40
That's that's reviewed by medical expert.
35:44
Obviously we want to make sure that that is reviewed correctly,
35:47
but but it really needs to be automatic.
35:50
So you sit down and and within 10 minutes you have a Neuralink
35:54
device installed very, very fast.
35:57
I mean, it's very sort of cyberpunk, you know, Deus Ex if
36:00
you played those games and. We'll Neuralink start to
36:06
interface with other devices like wheelchair.
36:08
It's great question. We currently focusing on
36:12
controlling computers and unlock independence in the virtual
36:17
world. Of course, our plan is as we
36:19
mentioned earlier, robotic arm and a wheelchair.
36:21
To unlock independence in the physical world is of course at
36:25
the additional risk if you make your computer.
36:27
There's some yeah to that, but we are working with the FDA to
36:32
allow us to be quite some. Well, it seems like if if the
36:35
wheelchair has a. An app.
36:37
Well, the wheelchair just need to have some have an interface.
36:39
It does, yeah. So wheelchair has a Bluetooth
36:42
interface. You could just Bluetooth
36:44
interface to the wheelchair and and that's probably something we
36:49
should do. We're working on pretty soon,
36:51
yeah. It's really a matter of
36:52
paperwork. I'm showing that you can do it
36:54
safely. You don't want to drive off.
36:55
A Cliff, Well, I think we could. Well, we can limit the speed.
36:58
Yeah. Yeah, exactly.
36:58
So it doesn't go careening often to disaster, but you know, so
37:03
just make it go slowly at first. But yeah, so being able to sort
37:08
of it really the New York device just should work generally for
37:12
anything that's got a Bluetooth. Interface, including potentially
37:17
an Optimus. Yes, yeah, you yes, you could
37:23
communicate with Optimus. Yep, absolutely Optimus.
37:29
We'll also be able to talk to Optimus.
37:31
But like, but you could just, yeah, instead of talking just,
37:37
you could just beam it directly. Or if, if someone has lost the
37:40
use of speech, then then they can still communicate to an
37:42
Optimist. They, they can communicate
37:45
telepathically to Optimists or by Bluetooth.
37:49
And, and, and so even if someone has, you know, completely less
37:54
the ability to speak, they could still control Optimist or their
37:57
computer or phone. I.
37:59
Mean also like if you have an optimist and you have a neuron,
38:01
you can just directly map the brain signal to control of the
38:03
physical arm of the robot. And that's a very meaningful
38:05
thing. Like if you're, you know, folks
38:07
that have spinal cord injury, one of the biggest requests is
38:09
to be able to scratch yourself. It's something that quite
38:12
annoying actually. And if you have a scratch on
38:15
your face, you can't fall asleep until you scratch it.
38:18
You know, it's very convenient to be able to move something
38:19
physically towards you, to be able to scratch similar things
38:22
like eating food. You know, if you need somebody
38:23
to feed you very hard to have dinner with friends in a way
38:26
that is, you know, sort of a normal social experience.
38:30
And so if you can feed yourself, pick up a fork and actually eat
38:33
a piece of chicken on your own, you know, that's a big deal.
38:37
It prevents and saves a lot of interactions with caretakers and
38:40
other people in your life that you rely on to take care of it,
38:43
But it really increases you. And I think an exciting
38:47
possibility long term also is to say if you take parts of the
38:50
Optimus Optimus humanoid robot and you combine that with a
38:53
neural link, let's say somebody has lost their arms or legs.
38:57
Well, we, we could actually attach an Optimus arm or Optimus
39:00
legs and do a neural link implant so that the, the motor
39:05
commands from your brain that go would go to your biological
39:09
arms. Now go to your robot arms or
39:10
robot legs. And again, you you'd have
39:13
basically cybernetic superpowers.
39:15
Actually, so the latency from the neural link to your hand
39:18
would probably be slightly faster than it is just to go to
39:21
your physical hand. So you can imagine like if
39:23
you're a piano player or AI don't know anything that
39:25
requires extremely fast hand movements, that you could
39:28
actually have a pretty imbalanced right hand robotic
39:30
arm control versus left hand physical arm control.
39:33
There's more. Yeah, like I said, it's just
39:36
kind of a cyberpunk Deus Ex in the future where you have
39:40
cybernetic upgrades that are actually better than your
39:43
biological limbs. And it's, it's certainly the,
39:51
we'll have a much, you know, as as particularly as we expand to
39:55
a large number of of, of customers or patients for neural
40:01
link, the understanding of the brain will improve dramatically
40:04
because really there isn't a fine, very fine grained
40:09
understanding of the brain today because the, it's just the
40:12
sensors aren't good enough. You've got fMRI, which is pretty
40:16
good, but it's still not as good as actually having high
40:19
bandwidth electrodes in the brain.
40:21
Yeah, I think this is under appreciated as a research tool
40:24
to to move that whole effort forward of really knowing, you
40:27
know, what the physical substance of human thought is.
40:32
We don't know to the to the degree that we need to.
40:35
So Neuralink is actually a very powerful research tool.
40:40
Yeah, I mean, we, I think we can ultimately understand and and
40:46
fix it's quite severe psychosis or like if somebody's got like
40:51
the if somebody's got like a. Like a delusion that they have a
40:56
chip in their brain. Yeah.
40:58
I was wondering if you were going to mention that one.
41:00
We just want to be clear. There's only one person with a
41:03
new linked chip in their brain. So for people out there who
41:07
think we've put a chip in their brain, we'd like to assure you
41:10
or what it's worth, you probably won't believe us, but we did not
41:13
put a chip in your brain. OK, So there's actually a
41:20
remarkable number of people who think we have put a chip in
41:23
their brain, but we have not. But in the future, if you would
41:26
like us to put a chip in your brain, which will perhaps help
41:30
with the issue of thinking that you have a chip in your brain,
41:33
then we will be able to do so. So there are people that have
41:37
severe schizophrenia. They've got basically things
41:40
that their brain is malfunctioning in some way.
41:43
And, and this is actually due to really like physical circuitry
41:49
issues. You can think of the brain as
41:50
like really it's a, it's a biological computer.
41:54
And if some of the circuits are crossed, it's going to, you
41:59
know, it's going to crash or it's going to have issues that
42:02
was not work. But with a Neuralink device, we
42:06
can fix those issues and, you know, give someone who I think
42:11
has to say severe schizophrenia or psychosis of some kind, allow
42:16
them to live a normal life. I think that is one of the
42:19
likely things in the future. So, yeah, I mean, yeah, you can
42:31
certainly imagine, like, I'm sure people have, like, parents,
42:34
grandparents who've, you know, have memory that's not working
42:41
as well as it used to be. Sometimes they forget who who
42:44
their grandchildren are or what day it is.
42:46
And this is something that on your linked device could help
42:49
fix I. Mean that that's actually one of
42:52
the personal reason in in in many way like forms of you're
43:02
you're literally losing your and then part of your right then,
43:04
which is a just a very, very. Yeah.
43:11
And it's really just, it's a glitch in the biological
43:13
computer that is a fixable glitch like, like it's a short
43:18
circuit essentially. How does the device charge and
43:22
how long does the charge last? Yeah.
43:24
So the current version that Nolan has, it lasts for four to
43:27
five hours on a single charge, and it takes about 45 minutes to
43:31
charge. The thing we've learned from
43:32
Nolan is that that's actually one of the main limiters for him
43:35
using it more. It's actually pretty hard to use
43:37
a product more than like 70 hours a week.
43:40
But that's about what he has he used it for in some weeks.
43:43
Yeah, 70 hours in a week, Yeah. I mean, just for context, like
43:45
you sleep roughly 8 hours a night.
43:47
So that's, you know, we're doing better than the bed.
43:49
The bed is 56 hours a week if you use roughly.
43:51
And so 70 hours a week if uses. I challenge you to think about
43:55
products that you've actually used for that duration.
43:57
But that's OK. Some of these points are worth
43:59
like emphasizing again, like the that Nolan, our first neural
44:03
link recipient, has used the neural link device for 70 hours
44:08
in a week, which is incredible. You probably won't enjoy the
44:11
time sharing his computer use publicly, but I mean, I assure
44:15
use for productive things only. But actually, so one of the
44:19
things we've learned is that in the next version of the device,
44:21
we really need to like double or you know, increase that battery
44:23
life. And so I think DJ, the next
44:25
version is going to be double. Actually actually double without
44:28
without increasing the charge. Correct in charging time to
44:32
double the battery life, meaning you should get roughly 8 hours
44:34
of use. And the goal is to actually get
44:36
to all they use so you can just charge, you know, maybe in your
44:39
sleep your. Sleeping pillow exactly as soon
44:41
as you've got like 16 hours of usage, then you basically have
44:44
24 hours of usage because it can charge while you're sleeping.
44:47
One of the things that's important, I think, to call out
44:49
here is if you're paralyzed, you can't, you know, put the charger
44:51
over your head yourself. And so it's important to think
44:53
about like it's not just a duration of better use, but also
44:56
can you recharge it yourself independently.
44:59
So we spend a lot of time thinking about how to make that
45:01
feasible because then that means that you can, this is what no
45:03
one does. You can use the device, charge
45:05
it, use the device, charge it, use the device without needing
45:08
anybody to come in and sort of help you with that, which is a
45:11
big deal if you're trying to play save until 5:00 AM at night
45:13
when your family's asleep. And the way in which he does
45:16
that is that there is a charger coil that's a big or you know
45:19
about this big. And we actually put it in the
45:23
sleeve of a, of a hat, a little beanie or a beanie.
45:26
And then he wears it and then says with the voice command and
45:29
charge. Charger energized.
45:31
That's the one he likes, yeah. How would writing work?
45:41
So so yeah, the current device that Nolan has is reading.
45:49
So it's trying to read his essentially like worst movement
45:52
from from one one hand. That's also, you know, with
45:56
pointing out like in the future, like I think we're pretty cool
46:00
to you can build into a second implant that would allow the
46:03
other hand to be used and also have higher, obviously higher
46:07
active electrode count. So then you can play 2
46:10
essentially play games 200 because that's normally how you
46:12
play games. And but then with with writing,
46:16
it's really just it's an electrical impulse instead of
46:19
like reading electrical impulses from the neuron to you issue an
46:22
electrical impulse, which is obviously critical for vision.
46:25
So vision is, is writing, which is just training and electrical
46:30
impulse in the vision part of the brain.
46:33
And that like activates a pixel. So we, we actually do have this
46:37
working in monkeys. So we had, we've had it working
46:39
with monkeys for a while now where you can sort of flash a
46:42
pixel and then you watch where the monkey, obviously the
46:46
monkey's like, what monkey's a little surprised to see like,
46:49
hey, there's a flash here and a flash here.
46:51
But it's gets used to it after a while.
46:54
But it just you can you can see that that the pixel is in the
46:57
right location because the monkey's eyes will dark to that
46:59
location. It's not on the screen, like
47:01
there's no pixel on the. Screen.
47:02
There's no pixel on the. Screen.
47:03
Yeah, just like. You just verify that that the
47:06
that you're triggering a pixel in the right part of the brain.
47:08
So, you know, the initial resolution for vision will be
47:13
relatively low, you know, sort of Atari graphics type of thing.
47:16
But over time, it could potentially be better than
47:19
normal vision. And then I guess in terms of
47:22
some additional applications for where writing to the brain can
47:27
be useful order applications as Bliss mentioned there is.
47:32
Feedback. There's a proprioceptive
47:33
feedback, there's a tactile feedback.
47:35
Especially for a robot arm, Like if you're trying to grasp a call
47:37
right, you need to know if you've got it.
47:39
Yeah, 1 to 1 egg. That's an egg, yeah.
47:42
It's a very much a delicate balance of not just initiating
47:45
the movement but getting the feedback and controlling it
47:48
accordingly. So there there is a some meta
47:50
sensory cortex that's right adjacent to motor cortex that
47:53
could could be beneficable. Motor movements, so any changes
48:07
in neural growth after device is inserted, we don't see any any
48:11
signs of neural damage. But I and I guess we, we have
48:16
seen some rebound on some of the electrodes, right?
48:23
Correct. And then also, I mean, I guess,
48:24
I guess, you know, rain is very plastic.
48:27
So it's not that plastic well. It does diminish quite a bit
48:31
after each and 20. Throughout childhood, actually,
48:39
when you get to about 25 rain really done cooking.
48:44
Yeah. But there are, there is a little
48:51
bit of damage done with each insertion, but it's a miniscule
48:54
amount compared to anything else out there.
48:57
And so it's an easy amount of damage recover from.
49:03
And it's really only detectable on cutting pieces of the brain
49:09
after after the animal's no longer alive and looking at them
49:13
under a microscope, you can't really tell during life that
49:15
there's been any brain. Good.
49:19
Another way to interpret this question, have there any changes
49:21
in neural growth after the device is inserted?
49:23
One way to interpret that is like the user learning how to
49:25
use the device. And I think on that side of
49:27
things, there's been tremendous progress.
49:29
He's put in hundreds of hours trying to figure out the best
49:33
way to use this device because he really thinks that, you know,
49:34
if he can figure this out, he can help share this knowledge.
49:38
I mean, he's like on Friday night at 8:00 PM, you know, he's
49:40
starting a session of like, you know, figuring out himself how
49:43
to how to push his own performance to the next level.
49:46
And that's really a unique learning process because there's
49:48
not many people in the world that had the experience of
49:50
moving something. And so there's a lot of nuance
49:53
to like, OK, how exactly should I imagine or attempt to move my
49:56
wrist to get the thing to. Yeah, he's really dialed that
49:59
into a. Also, just the sheer number of
50:02
hours that he's using even in the past six months, right?
50:06
Yeah, in many ways. Like, I mean, he's using it in
50:09
his travel and in his plane, right?
50:11
Effectively, PCI has left the lab, yeah.
50:13
Yeah, one of the questions is how close are we converting
50:16
thoughts into text? I mean, right, Right now it's
50:19
more about moving curse from the screen on on a virtual keyboard,
50:22
but long term you should be able to really estimate entire words
50:27
faster than anyone could possibly type.
50:30
I'm able to type Hello World today, but we're still in the
50:34
early days making that a posh experience.
50:36
I. Mean the other things that we're
50:39
looking at is sign language, right?
50:40
At the end of the day, it is a movement of and into, right?
50:45
Yeah, it's true. Was the brain trying to
50:49
naturally push the threads out? I mean, this is sort of a
50:52
universal feature of any implant in the body.
50:54
The body tries to reject it. And the goal of the surgeons and
50:58
the technology team is to fight that.
51:00
And so with artificial hips and with, you know, screws in the
51:04
spine, we've done a really good job of finding biocompatible
51:08
materials and techniques to fix those implants in the body.
51:13
I mean, past a certain age, it's getting hard to find someone
51:16
without some kind of implant, you know, knee.
51:19
Hip. Some kind of screws in their
51:21
spine. And so we've got this problem
51:25
pretty well solved. So to answer your question, yes,
51:29
the body is trying to get rid of any implant, but we can ensure
51:33
that basically can't. It's also worth highlighting
51:37
that the threads have not actually moved in the past five
51:42
months. There's, there's some still
51:44
minor movements in terms of like some maybe maybe getting pushed
51:46
in a little bit, pushed out a little bit, but it's, it's more
51:49
or less very stable and been stable for.
51:52
And the reason for that is, you know, once you, once you do a
51:56
brain surgery, it takes some time for the tissues to come in
51:59
and then, and then you know, our tissue or the neo membrane to
52:02
actually come in and then anchor the threats in place.
52:05
And once that happens, everything has been stable and
52:08
seen much movement. That's where the world record
52:10
performance starts to come. In Yeah, that was a couple weeks
52:13
ago. Yeah, the threats, like it is
52:18
important that the threats be extremely tiny.
52:20
If they're extremely tiny, then the brain does not.
52:23
The smaller they are, the less likely the brain is to react to
52:26
to them. So that's why you want the
52:28
threats to be extremely tiny and also to minimize any damage to
52:31
neurons, so. On that note, we do plan to
52:37
actually share some of the, you know, the tissue response in
52:41
detail and some of the the later upcoming updates.
52:44
Yeah, it is quite a challenging, it's challenging on many fronts
52:49
to do something like this because you're, you're trying to
52:52
read and read and write electrical signals, but you need
52:56
to have the, the threads themselves need to be like
53:01
electrically isolated and and not subject to corrosion in the
53:06
body. So like the, you know, just
53:09
metal by itself is somewhat subject to corrosion or, or
53:13
being attacked. So it's it's, it's like in terms
53:17
of the various coatings and things to actually make this
53:19
electrode work while not actually eroding its performance
53:23
over time is, is very difficult. Human bodies are very, very
53:26
harsh environment, very harsh environment.
53:27
It's a it's a bag of salt water with bad sensors that's elevated
53:31
temperature that is well regulated.
53:33
I mean, I'm sure people have experienced dropping their
53:35
electronic devices in a sea water and in an instant.
53:38
Yeah, yeah. So we'll sort of wrap wrap this
53:43
up soon if there's like a few few last questions, I guess.
53:49
So a good question. So what about upgrades?
53:52
So yeah, we do think it's going to be important to be able to
53:55
upgrade the device over time, just like you wouldn't want like
53:59
an iPhone 1 stuck in your brain forever.
54:02
You know, if you've got an iPhone 15, you probably want the
54:05
iPhone 15, not the iPhone 1. So I think people over time will
54:13
be able to upgrade their neural link.
54:14
So we'll take the neural link device out and put a new one in.
54:19
And we have done this with some of our animals and they're
54:24
actually in one case we did with we upgraded device three times
54:28
and so with a pig. We did with a monkey as well.
54:32
Monkey is able to do PCI. Yeah, and he's doing fine.
54:36
Pager has the latest implant. Actually, he hit his, I think
54:39
his record with the last, yeah. With the with an upgrade.
54:42
No, it still beat him though Will.
54:44
It still beat him. Yes, this is true.
54:46
Humans are top of the species leaderboard right now.
54:48
Pager's like what? Like 8 or something.
54:49
Pager's like 8.5 BPS, OK, but it's a very high score, yeah.
54:53
I'm not trying to put Pager down.
54:55
And also to train a monkey to do that, It's like a whole
54:57
challenge on it's own. We have like the best animal
54:58
care team world. Yeah, I just do 1/2 size.
55:01
We, we, we do our absolute best to take care of the animals.
55:06
And when we had like a USDA inspector come through, she said
55:10
that this was the the nicest animal facility she's ever seen
55:15
in her entire life. So.
55:17
Breakfast on an app like. The the the monkey orders room
55:22
service. Yeah.
55:24
We're, we're, we're monkey room service, which is rare.
55:28
Like we're the only ones who offer monkey room service.
55:31
So we really do everything we can to maximize the welfare of
55:35
the animals. So all right, with that, thank
55:38
you everyone for tuning in. Hope you found this interesting.