Latest / The Starting Block / Breaking Down AI’s Threat to Cryptography /w Kostas Chalkias

Transcript

Breaking Down AI’s Threat to Cryptography /w Kostas Chalkias · The Starting Block

0:11
Good morning, good afternoon,
0:12
wherever you are joining us from
0:14
today, welcome to the starting
0:15
block.
0:16
It's a Friday, so that's one thing
0:17
to be happy about.
0:18
And if you look above me in the
0:20
price ticker, that's another thing
0:21
to happy about, Bitcoin nearly
0:23
hit $80,000
0:26
for the first time in a while.
0:29
And yeah, it's looking green across
0:31
the board. So it's been an exciting
0:32
24 hours or more
0:34
for the overall cryptocurrency
0:36
market. And you can definitely feel
0:38
a palpable, we are back.
0:40
Across crypto Twitter.
0:41
So I'm loving that today, feeling
0:43
very positive,
0:45
been really interesting,
0:47
few hours on Twitter as
0:48
well, seeing everyone's reactions
0:50
and breakdowns of what
0:52
is really leading this big
0:54
recovery across the markets.
0:56
And it's great to see a little bit
0:57
of Hopium right before the weekend,
0:59
what's there not to be excited
1:01
about. If you're joining us for the
1:02
first time, welcome to the show.
1:04
Today is officially us
1:06
wrapping up two months of the
1:08
starting block.
1:09
Crazy to think more than eight weeks
1:10
ago we were in Abu Dhabi sitting
1:12
alongside CZ live
1:14
launching the show.
1:16
If you missed that conversation you
1:17
can go onto the Blocks YouTube
1:18
channel and find our interview with
1:20
CZ and a number of
1:22
other big names in the space Charles
1:23
Hoskinson, Paolo Arduino,
1:27
many many other big name in the
1:28
industry that we've had the pleasure
1:30
of speaking to on the starting block
1:32
in the last few weeks.
1:32
So go and check those out on our
1:34
YouTube channel If you don't
1:35
already, please subscribe.
1:37
And if you're watching us on X
1:39
today, go ahead and like and retweet
1:41
this broadcast so we can get more
1:42
people joining us for today's
1:44
conversation. A little bit later on
1:46
the show, we're gonna be joined by
1:47
Costas Chalkias, who's of course one
1:49
of the co-founders of Mistin
1:51
Labs, a real OG in the
1:53
cryptography space as well.
1:55
So we're going to have a really
1:56
fascinating conversation today
1:57
around AI, cryptography
1:59
and the general industry.
2:01
So really looking forward to getting
2:02
him on screen with us
2:04
a little bit late on the Show.
2:05
But of course.
2:06
We always start the show with the
2:08
block briefing, which is our
2:09
breakdown of top stories
2:11
you need to know in the last 24
2:13
hours. This is, of course, curated
2:15
from news that our
2:16
exceptional editorial team is
2:18
putting out every day.
2:19
So if you don't already, bookmark
2:21
theblock.co, our website,
2:23
on your browser.
2:24
Go and check out our news.
2:25
We also have a mobile app if you
2:27
want to stay more up to date.
2:29
And, of, course, if you follow us
2:30
already on X, please hit that Follow
2:32
button. And again, like and retweet
2:34
this broadcast so we can have more
2:36
people joining us today.
2:38
But I'm gonna jump straight into the
2:39
news so we get to the big
2:40
conversation of the day.
2:42
And starting off with our first
2:43
headline, breaking on the DeFi
2:45
front.
2:47
Mantra slammed the brakes on its
2:49
entire network earlier today.
2:51
The real world asset layer one froze
2:53
all transactions after spotting
2:55
an incident inside its Cosmos
2:57
EVM module telling users on
2:59
exit everything was paused while the
3:01
team investigated.
3:02
Here's the good news.
3:03
Mantra says it has since contained
3:05
that threat, traced the root cause
3:07
to just two wallet addresses
3:09
and crucially no user
3:11
funds have been exploited.
3:13
That chain could be up within
3:15
a day and we'll be keeping an eye
3:16
out from updates from the team.
3:18
This is, of course, pending testing.
3:21
And for a bit of history here, the
3:22
OM token though took another
3:24
hit down nearly 10% in
3:26
24 hours.
3:27
And this is, of course, after.
3:29
Mantra collapsed back in
3:31
2025, which was widely covered
3:33
in the industry.
3:34
I also had a conversation with JP
3:36
Mullin, the CEO at the time,
3:38
in the hours after that incident.
3:41
So not good news for Mantra once
3:43
again, and we'll be keeping
3:45
tabs on that story as it
3:47
develops.
3:48
Then onto markets, Bitcoin is on a
3:50
tear, pushing as high as $27,000,
3:52
$5,000 on Friday before
3:54
easing back towards $78,000.
3:58
Bernstein analysts think this could
3:59
be a real momentum shift.
4:01
And the thesis here is all about
4:03
liquidity.
4:04
The U.S. Treasury is doubling its
4:05
buybacks of longer data debt.
4:08
And Bitcoin has always loved loose
4:10
liquidity.
4:11
Meanwhile, the ETF story has flipped
4:13
spot. Bitcoin funds pulled in
4:15
more than one point six billion
4:16
dollars this week with BlackRock's
4:18
iBit leading the charge and
4:21
strategies back in the black on its
4:22
eight hundred forty thousand Bitcoin
4:25
stack.
4:26
Ether is actually outrunning
4:27
Bitcoin on this leg
4:29
higher. So really, really
4:31
interesting day across the
4:32
cryptocurrency markets, Bitcoin on
4:34
an absolute tear.
4:35
We nearly got to 80K and
4:37
seeing lots of takes on this really
4:39
being driven by that US
4:41
Treasury News in the last couple of
4:43
days, doubling down
4:44
on those buybacks of
4:46
debt.
4:48
So gonna be interesting to see how
4:49
things progress into the weekend,
4:50
but right now looking very much
4:52
green above my head on screen.
4:54
Then to Washington, a rare peek
4:56
inside the Oval Office.
4:57
President Trump reportedly huddled
4:59
with the who's who of crypto on Wall
5:01
Street, including Coinbase,
5:03
Kraken, Ripple, Robinhood, NASDAQ,
5:05
and ICE.
5:06
One source told the block's news
5:08
team that Trump was, quote, bullish
5:10
on getting the Clarity Act done.
5:12
Chainlink co-founder Sergey Nazarov,
5:14
who was in the room, called a
5:15
productive, mostly soliciting
5:17
feedback on what it'll take to pass
5:19
this bill.
5:20
A strategic Bitcoin reserve even
5:21
came up, though no next steps.
5:23
Have been explained
5:26
upon. The first Senate procedural
5:28
vote is set for September 15th
5:30
for clarity, but the bill is still
5:32
tangled in an ethics fight over
5:34
Trump's own crypto wealth
5:36
and interests.
5:38
Then it got spicy over at the CFTC
5:40
during an advisory committee meeting
5:41
yesterday.
5:42
CME Chief Terrence Duffy warned that
5:44
some prediction market contracts are
5:46
ripe for manipulation,
5:48
quote, horrible for our industry.
5:51
CFTC chair Michael Selig
5:53
cut him off, snapping those
5:54
examples, that those examples
5:56
were offshore and quote,
5:58
fake news.
5:59
Then Kalshi's COO
6:01
Luana Lopez Lara jumped in asking
6:04
if CME had ever
6:05
had a manipulation problem.
6:07
Duffy fired back here that he has
6:09
more people in his regulatory
6:10
department than her entire company.
6:13
Her reply, maybe you should learn
6:15
a little bit about efficiency.
6:17
Duffy then came back, maybe you
6:18
shouldn't learn about credible
6:20
markets.
6:20
So the turf war is clearly on
6:23
over who polices prediction markets
6:25
in the United States.
6:26
And I think it's a story we're gonna
6:27
be talking about for months and
6:29
months to come.
6:30
And Kraken wants to be your bank,
6:32
just not in America at least.
6:34
Parent company Payward is exploring
6:36
full banking licenses overseas.
6:39
Co-CEO Dave Ripley told the block
6:41
at the Wyoming Blockchain Symposium.
6:43
He sat down with my colleague,
6:44
Calvin Sparks, yesterday.
6:46
And we will be producing that video
6:48
and having it up on our YouTube
6:49
channel. And our X account in
6:52
the next 24 hours or so.
6:53
So keep your eyes out on that.
6:55
The pitch here, trading, banking,
6:57
and asset management, all under one
6:59
roof at Kraken.
7:00
Payward already runs Kraken
7:02
Financial, a Wyoming chartered
7:04
institution that recently snagged
7:06
a limited federal reserve master
7:07
account, a first for a crypto
7:09
firm, though it still can't lend
7:11
fiat or offer FDIC insurance.
7:14
And looking further out, executives
7:15
say they'd even like to offer
7:17
customers mortgages one
7:19
day. So.
7:20
Clearly seeing the evolution of
7:23
companies that started in the crypto
7:24
industry moving closer and closer
7:26
towards full stream financial
7:28
services that big
7:30
TradFi companies have been
7:32
offering for decades.
7:34
And that is it for the Block
7:35
Briefing this morning.
7:36
If you want to check out more news
7:37
relating to the industry, please do
7:39
go ahead and check out the Blocks
7:41
website that's at the Block Co.
7:42
We've got some fantastic journalists
7:44
who have their finger on the pulse
7:45
every day.
7:46
We've also got a couple of
7:48
newsletters that you can subscribe
7:49
to. On our website.
7:51
So I do encourage you to go and
7:52
check out the Blocks website if
7:53
you're interested in keeping up to
7:55
tabs with what's happening in the
7:56
industry. But that is it for the
7:58
news. We're gonna be moving on to
7:59
the hot seat.
8:00
It's a great pleasure to welcome
8:02
Costas Chalkias, who's one
8:04
of the co-founders of Mistin Labs.
8:06
We actually met in person a couple
8:08
of months ago in Paris at
8:10
Proof of Talk.
8:11
So first things first, Costas,
8:12
welcome back to the show.
8:14
It's good to see you again.
8:15
How are you doing?
8:16
Hey Gareth, good to see you guys,
8:18
good to see everybody.
8:20
Completely different environment
8:21
even since the Paris event.
8:23
As we can see, how AI is evolving.
8:25
Quantum is coming into the picture,
8:27
of course, the last few months.
8:30
I think we have a lot of things to
8:31
see this year.
8:32
Actually, 2026, at least in
8:34
my brain, is probably something
8:36
of a change in era
8:38
of AI and crypto and
8:40
general security.
8:42
This is something I'm.
8:44
I'm very curious to learn more and
8:46
how I can even contribute myself.
8:47
You know my background, Gareth,
8:49
right? I'm in this space forever.
8:51
My hacking abilities have been
8:53
diminished.
8:55
Just because I have a competitor
8:56
now.
8:57
That's the reality.
8:59
Of course, you need to use it
9:00
properly. But yeah, here we are.
9:02
Let's discuss on what's happening
9:03
today. What is the risks?
9:06
Ask me very difficult questions.
9:08
I'm someone who is investing a lot
9:10
of time since the very past
9:11
from the insurability of all of the
9:13
blockchains.
9:15
And now it's probably the time for
9:16
everybody to understand the risks
9:17
and also what we're doing as
9:19
cryptographers to protect your
9:20
assets.
9:21
Yeah, sure. I think let's just start
9:23
there, Kostas.
9:24
Obviously, the DeFi sector in
9:26
2026 has been in the crosshairs
9:28
with a number of big exploits,
9:30
but lots of experts have been
9:32
telling us that these are not smart
9:34
contract exploits.
9:35
But if we look more recently at
9:37
incidents like the cold card hack,
9:39
which is really a look at
9:41
their systems and
9:44
people using AI tools to
9:46
go and vulnerabilities across
9:48
the internet, across software
9:50
systems.
9:51
Really does seem that we're at this
9:53
very dangerous point where the
9:55
power of the AI tools that we have
9:57
are far more
9:59
consequential than we may have
10:01
realized and everything might
10:03
be at risk. And right now it seems
10:04
like crypto is in the crosshairs,
10:05
but that could be the entire
10:07
internet.
10:08
What's your general sentiment right
10:10
now on the power of these tools and
10:12
the risk that they pose to
10:14
our industry and the broader
10:15
industry? Or do you think that I'm a
10:17
little bit too alarmist and maybe
10:19
blowing it out of proportion?
10:21
You're totally actually on the
10:23
point. The reality is a few years
10:25
ago, we thought that crypto might be
10:26
broken by quantum computer, right?
10:28
You remember everybody started in
10:30
2022, 23, even Satoshi
10:32
at the very beginning mentioned one
10:33
day, maybe we have to switch to
10:35
something else.
10:36
But then we gradually realized we
10:37
have a new species in our world.
10:39
And this new species is actually
10:40
something that is even smarter than
10:42
us. I wouldn't say necessarily on
10:44
everything, but it definitely scales
10:46
more than the human brain.
10:48
And the whole idea and what you see,
10:50
I believe the
10:52
important date for
10:54
humanity was actually December 2025.
10:57
And obviously, everything adds up
10:58
into 2026, where for the
11:00
first time we realized with all of
11:02
this agent
11:05
coordination and a few other things
11:07
that you can automate search on
11:09
random stuff across the world, you
11:11
don't need to be a hacker anymore.
11:12
You don't have to be the security
11:13
expert.
11:14
People are just using it for
11:16
fun or some more experts, including
11:18
North Korea and a few other
11:20
organizations, even
11:22
on the dark web.
11:23
Now they have a weapon in their
11:24
hands and obviously they use it
11:26
without even them knowing what it
11:27
will discover.
11:29
This was not the case in
11:31
the far past. I would say that 2023,
11:33
2024, even at
11:35
Mist and we invented some,
11:37
discovered a few algorithms that
11:38
could use the early versions of
11:39
ChatGPT back then.
11:40
We call the project Bagdar.
11:42
We found a few situations like
11:44
scanning GitHub.
11:45
We realized that if you even defy
11:47
platforms outside sui, they had
11:49
their secret keys in their test
11:51
net for some reason.
11:52
And it was by far easier now to do
11:54
all of this scanning without the
11:55
human just opening the files, then
11:57
you create some automated scripts.
11:59
Now you have a personal assistant
12:01
and this is where we are, right?
12:03
These in the hands of someone who is
12:05
not an expert is already a weapon in
12:07
the hand of someone who is already
12:09
a white hat hacker
12:11
or a black hat hacker.
12:12
It's literally a nuclear bomb.
12:15
And to those who are telling
12:17
me sometimes, do you still
12:19
believe quantum is a thing?
12:20
We see what's happening with cold
12:22
card. We see, what's happened with
12:23
all of the random news you
12:25
hear every now and then, even
12:26
mantra that today's situation,
12:28
we don't know if it was a human who
12:30
discovered it, right?
12:31
Probably someone, even a security
12:33
guy, found bugged you
12:35
some AI automation testing,
12:37
and then obviously they had to
12:38
protect against this and so on.
12:40
I don't believe any attack anymore
12:42
is happening just with human brain.
12:44
If you ask my like position,
12:46
even if you are one of the best
12:47
hackers in the world, you still need
12:49
this help because you have an army
12:50
of personal assistants.
12:52
Now.
12:54
The cold card in particular is
12:56
probably the perfect example because
12:57
what we realized in this situation
12:59
and I want our people actually to
13:01
focus on this.
13:02
There was a bug like
13:04
for a few years.
13:06
It was already there.
13:07
Many people were looking at it but
13:09
nobody probably spent some attention
13:11
into automating and having a
13:13
continuous audit on what's happening
13:14
like with the old software.
13:17
Cold card is probably the best
13:18
example we realized that we cannot
13:19
stop auditing.
13:21
I think everything has to change.
13:23
From typically audit,
13:26
SIP, and then you forget into
13:28
continuous audits.
13:29
We cannot forget this.
13:30
Any new model that comes to our
13:31
hands, in my opinion, should be a
13:33
small crisis for everybody who's
13:35
working on the security industry.
13:36
And they have to run all of their
13:38
new models against the previous
13:39
code. In my opinion if the team
13:41
behind it actually has done it
13:42
before, unless it's something
13:44
internal like inside their
13:46
attack and so on.
13:48
With the new models, they should go
13:50
back. To all of the open source
13:52
code and this is a debate between,
13:54
you know, ledger of the closed
13:55
source information against Trezor
13:57
and Coldcard and all of these open
13:58
source wallets.
13:59
What is better these days?
14:01
If you ask me for the first time as
14:02
a cryptographer, I believe that
14:04
obfuscation in some situation might
14:06
be a difference because now you
14:08
have a security engineer who is
14:10
looking at all of the doors one by
14:11
one figuring out if you have an
14:13
attack, but the attacker, what do
14:14
they need, right? One door to be
14:16
opened.
14:17
And this is the big change, in
14:19
my opinion, in 2026.
14:21
For the first time, we realized that
14:22
if you don't do automation
14:23
internally, and I'm talking from the
14:25
different side, I don't care about
14:27
what the Black Hat hacker will do,
14:29
because they will try everything.
14:31
We need to automate everything,
14:34
not only on the new launches,
14:36
but everything in the past.
14:37
Gold Card is the perfect example
14:39
here.
14:40
Every substantial improvement in
14:41
frontier models effectively requires
14:42
you to do audits.
14:44
That's all, right?
14:45
When you see anthropic...
14:46
Publishing something, figure out if
14:48
you have some connection, you can
14:49
have a pre-release.
14:50
If you don't, the first day you have
14:52
to have a small crisis,
14:54
a small like war room to figure
14:55
out if your affected by, I don't
14:57
know, some particular prompt or
14:59
skills. This is my personal view and
15:00
I believe it will surpass all of the
15:02
attacks we will see from the quantum
15:03
computers.
15:04
AI is actually a bigger threat
15:06
to security and
15:08
if coins,
15:10
blockchains, DeFi platforms do
15:12
not have a security in at least one
15:14
person who is using AI to
15:16
defend. In my opinion, they are
15:17
making a big mistake.
15:18
Even myself as an investor, one
15:20
of the things, because yes, I'm
15:21
working on sui, I worked on
15:24
Ethereum as an auditor, in
15:26
other blockchains in the past, I'm
15:27
in the Bitcoin space since ever.
15:30
It's one of first question I'm
15:31
asking. Do you have a red team?
15:32
Do you a security guy who
15:34
is actually checking?
15:35
Because I want to put some money,
15:37
even if it's like 1,000 or 1
15:38
million, I need to ensure that you
15:40
have some protection against these
15:41
AI attacks, because this is the
15:43
biggest threat at the moment.
15:44
Let's face it.
15:45
It really is.
15:47
And I mean, I think you need look,
15:49
as you've said, no further than
15:51
the cold card incident.
15:53
And the response from the Bitcoin
15:54
community has been amazing.
15:55
I'm sure you're well aware of the
15:57
Bitcoin Red team that's formed.
15:59
And they've gone about using
16:01
Frontier models,
16:03
Kimi's best model and gone and
16:05
looked at all the Bitcoin open
16:06
source code, found as many
16:08
exploits or vulnerabilities
16:10
as possible, disclosed those.
16:12
He finds things.
16:13
Yeah, they find lots of things.
16:16
Which is really scary.
16:18
Something triggered at least our
16:20
community to actually be more
16:22
serious and sensitive
16:24
on this. Actually, it's good.
16:25
You know, in all of the innovations
16:27
that humanity did from the nuclear
16:28
energy and everything, we have a
16:30
small period where it's not a grace
16:32
period necessarily, it's a period
16:33
where this is exploited.
16:34
And then we need to figure out how
16:36
we can defend against this, then
16:37
eventually would be a balance.
16:38
And in my opinion, we're progressing
16:40
as a field
16:43
not only substantially,
16:45
but in my opinion in a huge way.
16:47
It says that we're learning to live
16:49
with this threat, right?
16:50
We have the wolf, we have the tiger
16:51
here, and now we try to create our
16:53
own differences.
16:54
I give one or two years, right,
16:56
one or three years for most of the
16:58
teams to realize without
17:00
the similar effort that the Bitcoin
17:01
community is doing, Sui as
17:03
well, we do have a specific rating
17:05
here.
17:06
You cannot move on, right.
17:08
It's very simple.
17:09
Don't invest in anything
17:11
if you don't have a security team
17:13
behind. Because previously it was
17:14
just sales, sales, and sales.
17:17
Now you realize insurance of
17:19
your network is actually more
17:20
important, otherwise you cannot
17:21
sell.
17:22
It's very simple.
17:24
Yeah, it makes me think to Zach
17:25
XBT's recent tweet, which was
17:27
actually eerily on
17:30
the money because I think it was
17:31
about two weeks before the cold card
17:32
incident happened where he was like,
17:35
hardware wallets are junk,
17:37
you can't trust these things.
17:39
And then two weeks later, the
17:40
cold-card incident happened.
17:42
And he was kind of suggesting that
17:43
you're better off using an
17:46
iPhone that you don't use anymore
17:47
that's a few years old because
17:49
of the closed source nature of
17:51
the code, with the security.
17:53
Would you argue that maybe
17:55
that is kind of true?
17:56
I have mixed feelings.
17:58
In my opinion, all of the world will
18:00
get into a multi-sig mode.
18:02
You need the hardware.
18:03
There is a reason you need hardware
18:04
because AI will also attack your
18:06
phone, it will also attack older
18:08
phones even easier than before.
18:11
We need the second factor of
18:12
indicator where you need proximity
18:14
to the actual victim
18:16
in order to steal the password or
18:18
whatever we're using for
18:19
authentication.
18:21
People are not, I think...
18:22
The early versions of blockchains
18:24
made, in my opinion, an important
18:27
mistake.
18:28
You need to charge the users
18:30
more just because they're using
18:32
multi-sig. And if you see what we
18:33
did at Sui, we erased that.
18:35
We don't penalize users because they
18:37
are using more complex cryptographic
18:39
authentication.
18:40
Because it's as if you're going to
18:42
Google and you say you're using
18:43
8-digit passwords or 30-digits
18:45
passwords where we need slightly
18:47
more hash invocations.
18:49
Google needs to pay slightly
18:51
more on energy.
18:53
You need to pay us.
18:55
It doesn't make sense.
18:56
We have to get into a mode where,
18:58
in my opinion, the hardware wallet
19:00
will actually be more important than
19:01
we believe. I'm against that
19:04
comment, by the way.
19:05
I believe you need both.
19:06
You need the old iPhone, of course,
19:08
but you need also to have a second
19:09
factor of indication because pure
19:11
software, especially on devices
19:13
that are old.
19:15
Are even more vulnerable to this
19:16
AI revolution, right?
19:18
This is my personal view as a
19:19
cryptographer.
19:20
I don't know if I'm giving advices
19:21
now that some people cannot do.
19:23
Figure out how to use
19:25
like wallets or put some pressure
19:27
to the communities like including
19:28
Bitcoin, Ethereum, Shui, as
19:30
ourselves, I'm not putting ourselves
19:32
outside the equation.
19:34
Guys, build very easy to
19:36
use multi-sig wallets with
19:38
some hardware support.
19:40
This is the only way to, you need
19:41
two factor authentication.
19:42
Like You cannot rely on anything
19:44
anymore.
19:46
And imagine myself, like I
19:48
live in Dubai, and imagine
19:49
everything is under camera
19:51
surveillance.
19:52
Sometimes you're even using
19:53
hardware.
19:54
There might be a camera behind you
19:56
stealing your password or stealing
19:57
your credentials, right?
19:59
We also have to be careful by
20:00
ourselves. It's like natural
20:02
defense. We see the attacks.
20:03
It's good that we have these
20:04
incidents, in my opinion, not the
20:06
end of the world.
20:07
Yes, a lot of money has been lost.
20:08
At the same time, There is progress
20:10
and we proved a lot of things before
20:12
something happened.
20:13
Imagine if Claude published Fable
20:15
or Mythos without telling anybody,
20:17
not having a pre-release, and this
20:19
came to the hands of North Korea for
20:21
one year, then it will be the end of
20:24
Web2 and Web3, right?
20:26
We have to do it cautiously, but
20:28
in my opinion, every individual
20:30
should also protect their keys and
20:32
ensure that they have some multi-seq
20:34
capability, even if it's more
20:35
expensive for other blockchains, and
20:37
so it will cost free.
20:39
Yeah, for sure.
20:39
And I think for most people,
20:41
it's just a big wake up call
20:43
to be very conscious
20:46
about the potential
20:49
mishaps of, you know,
20:50
having one single point of failure,
20:52
you know, so if it's a hardware
20:54
wallet and your seed phrase like
20:56
that, you're
20:59
playing a dangerous game.
21:00
Yes, you are safe.
21:01
Go ahead, Costas.
21:02
Gareth, my biggest fear here,
21:04
it's not only web3.
21:06
Have you heard some news that
21:07
someone actually lost the
21:09
seed, the mnemonic, because they
21:11
had it on Gmail and their
21:13
email was hacked?
21:14
And obviously, agents now, because
21:16
previously human, even if you could
21:18
attack an email, you needed some
21:20
scripts, you need so many other
21:21
things to understand, okay, where
21:22
could I find the mnmonic if someone,
21:25
I don't know, make some
21:26
anagrams on how they refer to
21:28
the mnomonic and so on.
21:30
Now it's very easy.
21:31
If I manage to get into your Gmail
21:33
account or even your Facebook
21:34
account, I can extract information
21:35
about you that I can use it either
21:37
for social engineering or literally
21:38
find the information inside your
21:40
inbox.
21:41
So we have to change that mentality,
21:43
not because we are talking about
21:44
Web3 and you're obviously
21:47
like me, an advocate of
21:49
this community. We need for
21:50
everything, for Web2 as well.
21:52
And this is, in my opinion, one
21:54
extra plus point I'm
21:56
giving personally to the
21:57
cryptographers and security
21:58
engineers in our space.
22:00
Because without us, and
22:02
if you see all of the big
22:03
innovations in the cryptography
22:04
space across the world, forget Web3,
22:07
most of them because of the money
22:08
that was in the blockchain space,
22:10
was because blockchain companies,
22:13
blockchain organizations, blockchain
22:14
foundations, had some of the
22:16
best cryptographers in the world.
22:18
And I can tell you for sure what
22:19
Mark Zuckerberg did in the Libra
22:21
days. You remember Facebook, for
22:22
those who don't know, a few years
22:24
ago tried to build another
22:26
blockchain internally, right?
22:27
It was called Libra.
22:29
It changed the name afterwards to
22:31
Diem.
22:32
But before that, Facebook didn't
22:34
have the amount of cryptographers.
22:35
You see what happened with Cambridge
22:36
Analytica and all of this.
22:38
Because of blockchain, however,
22:40
Facebook needed to hire and transfer
22:41
a few people across the globe, like
22:43
even myself. I moved from London to
22:45
the United States and it created
22:47
some of the best teams in the world
22:48
for security and cryptography.
22:51
All of these teams are now working,
22:52
I don't know, either for A16Z, some
22:54
VCs who are also investing in
22:56
crypto. Some other like companies,
22:58
some of them went back to NSA and
23:00
NIST and so on, but it started
23:02
with blockchain research.
23:03
We cannot, in my opinion, say
23:05
that blockchain is only like for
23:07
transferring money and maybe some
23:08
people don't believe in it.
23:09
On science side,
23:11
and the security of everything, our
23:13
AI agents, our emails and
23:15
everything, the blockchain
23:17
improvements on security
23:19
and cryptography was probably
23:20
fundamental for us to
23:22
defend against the new threat which
23:23
is AI.
23:24
Not only for our crypto, for
23:26
everything, right?
23:27
For your social life, for every
23:28
thing. And this is one of the
23:29
reasons that keep me busy
23:31
and engaged with the
23:33
sector. Otherwise, some people say,
23:35
oh, I will move from web threat to
23:36
AI because I don't trust the sector
23:38
anymore. I would say,
23:40
you will see that AI will definitely
23:42
need verified agent
23:44
execution very quickly,
23:46
vulnerability detection against AI
23:48
threats. Where is the red button
23:50
that will stop AI from running and
23:52
conquering the world, right.
23:53
This will happen from
23:54
cryptographers. Most of the
23:55
cryptographers live in the
23:56
blockchain world.
23:57
Let's face it, that's the reality,
23:59
right? Without this, you wouldn't
24:00
have this expert to defend against
24:02
AI. And it's one of the reasons that
24:04
I want crypto to survive because of
24:05
this.
24:05
Yeah Yeah, for sure.
24:07
And I mean, like you, I look at
24:09
people like you guys like Ilya
24:10
Polosukhin at NIA, who's, you
24:12
know, one of a foundational AI
24:14
engineer, who's now, you know,
24:16
everything is going back that way.
24:17
And exactly conversation we had with
24:19
him a few weeks ago was exactly
24:21
this, you know, we've solved a lot
24:23
of the problems that, you
24:25
know, and tackled a lot of the
24:26
problem that the AI and
24:28
the, like, general sort of
24:31
technology sector is dealing with.
24:34
So we're like purpose built for
24:35
this. I did want to speak to you
24:37
just very quickly.
24:38
Obviously, Sam Blackshear
24:40
announced that he was leaving Mistin
24:42
and going to Anthropic, which is a
24:44
big deal, right?
24:45
But it kind of speaks volumes
24:47
of the category
24:49
of people that were working at
24:50
Mistin Labs that are now going to a
24:52
company like Anthropic to help on
24:54
defensive security research.
24:58
Maybe you can talk to us a little
24:59
about that, like Sam's
25:01
contribution to Mistin and what
25:03
he's going to be doing there if you
25:04
do that.
25:04
By the way, for those who don't know,
25:05
Sam was the reason that I eventually
25:07
ended up into Silicon Valley because
25:09
he saw me in a conference in
25:11
Canada back in 2018,
25:15
I think, that I had the
25:17
paper with Mike Hahn.
25:18
Mike Hahn was the creator of
25:19
BitcoinJ in
25:22
the Satoshi protocol and the last
25:23
one who received an email from
25:24
Satoshi, he was my manager.
25:26
The reason I'm in the Bitcoin space
25:27
forever was probably like
25:29
potentially and partially because of
25:31
Mike. And then Sam invited me to
25:32
Facebook and I have very good
25:34
connection with Sam.
25:34
I know exactly how he's thinking.
25:36
He helped me a lot in my career and
25:38
he's probably one of the best
25:40
software engineers we have in
25:43
the Web3 space.
25:44
We still have, I didn't say had, and
25:45
I will explain why.
25:47
Obviously this is a personal career
25:48
decision for Sam, but don't forget
25:49
that this opens the door for crypto
25:51
companies to speak with real AI
25:54
foundational LLM companies.
25:55
Sam is still an advisor to Mr.
25:57
Nsui and there is a reason because
25:59
he's very, very sensitive.
26:01
On the security-like
26:03
defenses against AI and how we
26:05
can use AI to
26:08
do all of the security surveillance
26:09
and everything.
26:11
I don't believe, to tell you the
26:12
truth, that this is actually a bad
26:14
thing, but it's mostly a good thing,
26:15
because even for ourselves, we
26:17
see a new door opening into
26:19
these companies.
26:20
You already know that
26:22
Mr. In particular had some good
26:23
connections with Google and a few of
26:25
the early Gemini-like
26:29
progress on vulnerability detection.
26:30
Now we have with Anthropic as well.
26:32
Sam is fundamentally a security
26:34
engineer and researcher.
26:35
This is what he did for our
26:36
space. He tried to create probably
26:39
one of the best languages after
26:40
Solidity that was focusing on
26:42
security, which is Move, right
26:44
on top of Rust.
26:45
Obviously, Rust is known for being a
26:47
secure language.
26:48
And Move is a Web 3 language
26:50
in particular, like an evolution of
26:52
Rust to make this happen.
26:53
In my opinion, the.
26:55
Like modern version of Solidity.
26:57
I see Solidity like, I know some
26:59
people will make fun of me,
27:01
but it's like the Cobol and Pascal
27:03
of Web3, like
27:04
the early programming languages of
27:06
the world. And then you need an
27:07
evolution. The next evolution that
27:09
we have at the moment and which is
27:10
the best, in
27:12
my opinion, effort is Move.
27:14
And some is Move, right?
27:15
There is a reason I believe that
27:17
Anthropic also hired Sam.
27:18
I don't know exactly how he
27:20
performed in the interviews.
27:21
By the way, He was not also immune
27:23
from interviews.
27:24
Don't people, people shouldn't
27:26
believe that because.
27:27
Yeah, yeah, no, no.
27:28
He got an interview. He got the real
27:29
interview actually.
27:32
And Sam is very sensitive, right?
27:33
And by the time he realized,
27:35
oh, Anthropic cares about me and
27:37
they want me to lead the few things
27:39
around security for with
27:41
AI and all of this stuff.
27:42
Obviously I'm still young.
27:43
He's younger than me, by the way.
27:45
He had to take a decision, but he's
27:47
staying in web three, like as I
27:49
said, as not only an advisor.
27:51
The Move Advocate and all of this,
27:52
which is the best language in the
27:54
world at the moment for these
27:55
things. You never say the goodbye,
27:57
right? So in
27:59
my opinion, that's a major thing.
28:01
I personally switch, obviously,
28:04
it hurts me when I have to run some
28:06
tasks that some had to do
28:08
internally.
28:09
Like as a quiddle, right, I have do
28:11
a lot of more things
28:13
under my table now and my umbrella.
28:16
But at the same time, for the...
28:18
Like for the first time, since I
28:20
remember myself, I do have a door to
28:22
Anthropic.
28:24
This is generally, if I have it,
28:27
imagine there is an ecosystem around
28:28
us. I'm also active in other
28:30
communities. As you see, I'm working
28:31
on quantum, not only for Sui,
28:33
but some of the quantum solutions we
28:35
had, it's applying to Solana, to
28:37
Near, and a few other places.
28:39
Imagine if you also have someone who
28:41
knows Web3 really well, and now it's
28:43
inside the nest of AI,
28:45
literally in the bee's nest, in
28:47
the hive.
28:48
So that's why I believe it's
28:49
actually a good outcome, and I have
28:51
to actually myself make more
28:53
out of this.
28:54
Not only accept that Sam opened
28:56
the door, let's use the door now,
28:57
let's go there.
28:59
I think it's excellent, honestly,
29:01
and I appreciate you being so candid
29:02
and open about talking about what it
29:04
actually means because I think a lot
29:06
of people are skirting around it.
29:07
And for me, it was like, this is
29:09
awesome.
29:10
Someone who's built one of the
29:12
newer blockchain
29:14
protocols that we have is
29:16
moving to Anthropic,
29:18
to one of three or
29:20
four AI companies in
29:22
the world. Like take
29:24
a bow, crypto industry, we can
29:26
be pretty proud of how far we've
29:28
come. And I wanted to take the
29:30
conversation in the direction of
29:31
Walrus, because I wasn't even that
29:33
aware that Walrus has kind of taken
29:35
a big direction change.
29:36
You've got Walrus memory now,
29:38
which is essentially, as far
29:40
as I understand, from a layman's
29:41
explanation, a protocol that
29:43
allows you to take all of your
29:45
history and all the things you've
29:46
been doing with one LLM
29:48
and take it across to another LLm,
29:50
which for me is a big
29:52
deal, right? Because someone might
29:54
be a chat GPT user for the
29:56
last year.
29:57
But then they realize that, hey, I
29:58
really like Claude. I really what I
30:00
can do with Claude code.
30:01
And I want to go and do it there.
30:03
But I've done everything on chat GPT
30:05
up until now. And it's all in the
30:06
open AI systems.
30:07
How do I pull that across to
30:09
Anthropic System?
30:10
Can you talk to us about how
30:12
this kind of came about
30:15
and what it solves for the industry
30:16
that we don't have?
30:18
This is actually one of my babies
30:19
here.
30:21
Obviously, having a
30:23
storage and data
30:24
portability network like Walrus,
30:26
this is the reason that we had to
30:28
build Walrus because most of the
30:29
blockchains can handle short
30:31
chunks of data, probably just
30:33
transactions.
30:34
What do you do when you have to deal
30:35
with larger data files and so
30:37
on?
30:38
And our civilization is becoming
30:40
more hungry for data.
30:41
We see it like AI, even AI needs
30:43
data.
30:44
And then we realized that for the
30:45
first time, and to tell you the
30:47
truth, something that triggered a
30:48
lot what we do on Walrus
30:53
is Trump's statement,
30:54
I'm European.
30:57
Back in the days, not a few months
30:59
ago, actually, where Europeans
31:01
or anybody who is not an
31:02
American resident, citizen,
31:04
doesn't have access to mythos and
31:06
fable.
31:06
I was shocked.
31:08
I'm not imagining, in my
31:10
opinion, because I'm also like,
31:12
you know, I'm raised.
31:15
At least business-wise and
31:16
technically-wise in the United
31:17
States, Silicon Valley and some
31:19
London.
31:20
Now Dubai, of course.
31:21
My early career, of course, was in
31:23
Europe, but I'm still like my family
31:24
is from Europe and so on.
31:26
I felt like Europe doesn't have a
31:27
chance to survive what's
31:29
happening, right? We don't have
31:30
anything apart from Mistral.
31:32
That's the only effort I could
31:33
probably remember.
31:36
And this is like miles
31:37
away from what's happening in the
31:39
West and then to Far East.
31:40
Like to China.
31:42
Open source versus closed source
31:43
model. This is what we have and
31:45
there is nothing in between.
31:46
So I said, look, geopolitically,
31:49
and the fact that I have many, many
31:51
friends, I'm also an angel investor
31:53
in some situations and I know
31:55
some of them are working with Sui as
31:56
well on verifiability.
31:58
There are three elements that we
32:00
need to protect our data against
32:01
things that we're predicting
32:03
are coming.
32:05
One of them is the need for people
32:07
because the LLM performance
32:09
is changing.
32:10
To migrate from one place to the
32:12
other, like from Gemini.
32:13
Now I realize, oh, the new model
32:15
from Claude is better.
32:16
I go back to Claude.
32:17
And then, oh Gemini is now better
32:18
again. Why should I be second
32:20
compared to my colleagues and my
32:22
competitors in the world?
32:23
I go to Gemini and
32:25
then, Oh, Tzadzipiti made the new
32:27
evolution and then DeepSeq and then
32:29
whoever.
32:29
We need to port data.
32:31
But if you have only one American
32:33
company doing that, then eventually
32:34
you are a hostage of a particular
32:36
nation or a particular situation
32:38
like a company.
32:40
That eventually has all of the data
32:41
from all of the LLMs and then they
32:43
can build their own model.
32:44
And I know how this thing works,
32:46
right? I mean, I used to work at
32:47
Facebook three years.
32:48
I was the lead cryptographer at
32:50
Facebook. It's super important to
32:51
use this data for your own
32:53
evolution, building your products
32:54
and so on. Now, the new trend
32:56
is to build new LLM models because
32:57
after we reach the state where
33:00
the LM models became, I don't know,
33:02
you're improving slightly, then you
33:03
need data. Data is the king, data is
33:05
the new money.
33:06
But because it's the new money, it's
33:09
also our personalities.
33:10
Like you're talking, I guess you,
33:11
Gareth, as well, more to
33:13
the chat GPT, Claude and
33:16
Gemini than your
33:18
partner, your friends, even
33:20
inside your mind anymore because
33:22
you have a personal assistant that
33:23
you can ask them any day, like at
33:25
the restroom, like when you're
33:27
sleeping, when you are even driving.
33:29
Now you have voice, like when your
33:31
interviewing anything.
33:32
So it has everything about you
33:34
and actually knows more than
33:36
yourself because your memory as a
33:37
human is limited.
33:39
You might be surprised that in a few
33:40
years, the memories that you're
33:42
using on your GPTs, on your
33:44
Gemini and Claude is going to
33:46
be more than you can handle
33:47
internally as a Human Being, right?
33:50
Which means that we need a way to
33:51
decentralize this and not give the
33:53
power to a particular entity.
33:55
And Walrus was obviously the
33:57
obvious solution.
33:59
It's decentralized.
34:00
In a way that the data is not
34:02
controlled by a particular continent
34:03
or nation or company,
34:05
right? These are the three things.
34:07
People are porting now because it's
34:09
a new trend, just because the
34:11
performance is moving.
34:12
Also the cost, now the cost is also
34:14
very important.
34:15
So these three factors require
34:17
someone to be the base,
34:21
not addicted to a particular
34:22
nation or country,
34:24
that you can move your data from one
34:26
place to the other. The only thing
34:27
we had is decentralization.
34:29
The only decentralization that is
34:31
working with multisig, with some
34:32
quantum resistance now
34:34
efforts and so on, it's Walrus,
34:36
there was nothing else.
34:37
So we took the decision that
34:39
Walrus itself will deal with
34:40
portability of data and AI
34:42
because this is a natural evolution
34:44
of how we started and where the
34:46
world is now.
34:47
Geopolitics, the requirement of
34:49
people moving from one place to the
34:51
other, LLMs, and the cost.
34:52
Sometimes you might live with
34:55
like less power, less performance,
34:56
but you cannot for the cost.
34:59
So you need to migrate to another
35:01
LLM that is cheaper than you,
35:02
right? The one that you used to
35:03
have. So I believe it was natural
35:05
evolution. I personally see it as
35:08
convergent evolution at the same
35:09
time, Darwinism here.
35:11
It's a Darwin evolution.
35:13
We had to do it.
35:14
We had do it, otherwise there
35:16
is nothing else at the moment that I
35:18
think we are one of the first
35:19
who actually take that step in
35:22
a decentralized form, not a
35:23
particular entity. It's an
35:24
organization, a foundation.
35:26
Let's see how this will go, but I
35:27
have high expectations.
35:28
Seats. It's literally part of my
35:30
umbrella projects.
35:31
You need cryptography here a lot.
35:34
You also need verifiability of your
35:35
data because you need to verify, oh,
35:37
there is nobody injecting
35:39
data that you shouldn't want,
35:40
because then it affects your brain,
35:42
it affects your decisions.
35:43
So you need cryptographic.
35:44
It's actually one of the best
35:46
examples of cryptography being used
35:48
from web 3 to protect
35:50
against AI now and
35:52
AI migration of data.
35:53
And data is the key.
35:55
Data is the king.
35:57
I wanted to just chat
35:59
to you and pull on that thread a
36:00
little bit more about, you know, I
36:02
think it's about a month ago now
36:04
when, you know Anthropic announced
36:05
that, yeah, they would be limiting
36:07
access to Fable and
36:10
Methos and it was crazy, right?
36:11
No one expected that.
36:13
Do you know that I talked to my
36:15
wife, Gareth, I talked my wife
36:16
the very first, the very next
36:18
moment.
36:19
Maybe we need to move back to the
36:20
United States.
36:22
Like I've got one of my children
36:23
coming in this is the first time
36:25
this has happened on stream
36:26
Welcome, welcome.
36:29
You got to go downstairs you own
36:30
lots of people are watching on the
36:32
internet this is my son samuel go
36:34
on go downstairs
36:37
sorry i've got my wife coming
36:39
together no no no i have two
36:43
give me a mic
36:45
I have to as well, do not take your
36:46
time, take your time.
36:49
There we go.
36:50
I've been waiting two months of
36:52
the show and finally my son is
36:54
coming to the studio and bomb
36:55
me.
36:58
This funny video of someone
36:59
interviewing and his kid actually
37:01
came inside with someone like
37:03
grabbing the key with force like
37:05
outside the door.
37:06
This is this is amazing.
37:08
I think it was a famous BBC
37:09
interview.
37:10
But let me land the question again.
37:12
I'll reset very quickly.
37:13
Obviously, when they made that
37:15
announcement, it made me realize
37:17
how centrally controlled
37:19
these companies are and how
37:21
centrically controlled these
37:23
tools are.
37:24
And everyone wants to have equitable
37:26
access to AI.
37:28
And it's in humanity's best interest
37:30
that that is the case, whether
37:32
or not there are nefarious people
37:33
that can use these tools or you
37:35
know, white hat versus black hat
37:37
kind of, you know approach to this.
37:39
And I don't think that there is a
37:41
wholesale decentralized
37:44
AI alternative.
37:45
I know like Tethers QVAC is trying
37:47
to solve for this.
37:48
I think NIA AI is probably leaning
37:50
in that direction.
37:51
There's probably a few others,
37:53
models of companies that I don't
37:54
know about right now.
37:55
But how important do you think it is
37:57
to have a decentralized alternative
37:59
in this space?
38:02
In my opinion, it will definitely,
38:05
whatever we do, unfortunately,
38:06
it will create a societal gap.
38:09
I cannot see how we can avoid this.
38:12
We're not ready as like as
38:15
I would say culturally,
38:17
ethically, and a few more people
38:19
like there is also this capitalism
38:20
and you have to make also money.
38:21
Otherwise, it's the same thing with
38:23
drugs, right?
38:24
Some people are criticizing them,
38:26
you know, medicine companies and
38:28
you know the pharma situation.
38:29
Oh, why do you charge for these
38:31
drugs so much?
38:32
It shouldn't cost you more than
38:34
like one dollar now to produce it.
38:35
But you had to invest so many
38:37
billions until you come to this
38:39
stage where it's quite complicated
38:41
what you do. It's the same thing
38:42
because the same actually statements
38:44
have happened for pharma situation.
38:46
I don't know if you remember, right,
38:47
Trump also mentioning about, oh,
38:49
why do we pay so much for drugs and
38:52
like medical things,
38:54
right? Why Europeans are paying less
38:55
and so on.
38:56
It's not different.
38:57
Like we see that culturally we're
38:59
ending up in these situations.
39:01
Where for the very beginning, when
39:02
we have a new innovation, there is
39:04
also this patting world and
39:06
some debates on who is going to
39:08
win this war, because eventually
39:09
it's money. Let's face it, these
39:11
companies are for profit.
39:12
Like if an OpenAI started for
39:14
non-profit, and then we had all of
39:15
these debates, everybody knows.
39:17
And then at the other side of the
39:18
planet, we're having this more open
39:20
source way.
39:21
We don't know exactly how
39:23
the open source launches are
39:25
happening or if the data is the most
39:27
important stuff, because.
39:28
Even if you have an open source
39:29
product, then probably have access
39:30
to data and then this data actually
39:32
feed your next evolution of
39:35
like a product that one day might
39:37
not be open source.
39:37
So we cannot trust anything.
39:39
Because we cannot trust anything and
39:41
the world like in a capitalistic
39:42
scenario works like this, especially
39:44
at the very beginning of any
39:46
innovation that humans had, we
39:48
have to face it.
39:49
We need to gradually build.
39:50
It's good at the very beginning to
39:52
have this competition and obviously
39:53
protect your your IPs.
39:56
But we're getting in a more mature
39:57
environment, like AI
39:59
is running for at least three, four
40:01
years now that is so powerful that
40:02
can change like humans'
40:05
behavior and way of thinking.
40:07
We need to start investing in
40:09
these things. It will take time.
40:11
I'm not, I cannot promise to
40:13
everybody that whatever we do, Nia
40:15
is doing, some other like
40:18
visionaries are trying to do for
40:19
decentralization of AI.
40:21
It's still not there.
40:22
That's the fact.
40:23
We are working on whatever
40:25
we can to make the transition easier
40:27
down the road.
40:29
Walrus Memory is doing all of these
40:30
things for data portability
40:32
between LLMs, which is super
40:34
important. I personally use it, I
40:35
personally think it's a problem we
40:37
have to solve, but it's part
40:39
of the problem. It's not only this.
40:40
What do you do with all of this
40:42
models and the data that these
40:44
companies are accumulating, not
40:46
how people are storing the data and
40:48
they don't tell you or even
40:49
injecting information that can drive
40:51
opinions?
40:52
These things will require
40:54
cryptography.
40:55
My goal, and I believe this was one
40:57
of the reasons that
40:59
people don't believe it, but I think
41:01
in the back of
41:03
Sam's mind was,
41:05
let me go to the nest,
41:07
figuring out how we're doing all of
41:09
this security and privacy for our
41:10
data.
41:11
Sam is a very sane man.
41:13
Sam is someone who cares
41:15
about humanity.
41:16
I know, he cared about me.
41:17
He didn't know about me, right?
41:19
He sees like...
41:21
I don't know if he's
41:24
a sapiosexual or sapio,
41:25
you know, all of this.
41:26
He likes intelligence, but in
41:28
all of his aspects of his life, he
41:30
wants to contribute to humanity.
41:33
And in my opinion, this is also
41:34
something that I believe a few
41:36
visionaries on Web3
41:38
I taking
41:40
this into consideration.
41:41
We want to be able to
41:43
figure out how we can offer
41:44
decentralization, step by
41:46
step. It's not something that will
41:47
be sold in two years.
41:49
We have to start from somewhere
41:50
though, right?
41:51
There will be a few years until
41:52
these companies feel quite safe,
41:53
even financially, to be be able
41:55
share their ideas and models and
41:57
move to the next step.
41:58
And the next that might be, I don't
41:59
know, biomedical, you know,
42:02
Zuckerberg's foundation, by the way,
42:04
they're working on AI for
42:06
pharma, for finding.
42:08
I don't know, biochemical stuff.
42:10
We only, at the
42:12
moment, focus on text
42:14
and probably some optimizations
42:16
on trading and other things.
42:19
But in my opinion, AI will be far
42:21
more than what we have today.
42:23
So we're going to solve one of the
42:24
problem, eventually going to the
42:26
next step, which is protect humanity
42:27
against diseases and other stuff.
42:29
In the meantime, we're preparing the
42:30
ground for decentralization of all
42:32
of these data.
42:33
It will take time because if you
42:34
don't give some time to these
42:35
companies, they will close down.
42:37
And, you know, there is this debate,
42:39
what do you do in this, like,
42:41
business world at the same time,
42:44
they cannot do it for free,
42:46
right? It's quite complicated,
42:47
unless it's a national effort.
42:50
Yeah. Probably what's happening in
42:53
Far East, maybe,
42:55
and who knows, right, then this is
42:56
not exactly, exactly the case.
42:58
Eventually, there's someone behind
42:59
it. So we need to be careful
43:01
and, in my opinion, prepare the
43:03
ground for this. And this is what
43:04
I'm personally doing, right.
43:05
Starting with this even suite.
43:07
You will see suey becoming by far
43:08
more agentic and by far more
43:10
sensitive into the privacy sector.
43:13
There is a reason because we realize
43:14
that data, data is now
43:16
more important than everything.
43:18
And yes, there is trading, there's
43:19
DeFi, but along with this, there is
43:20
also data that is accumulated and
43:22
the blockchain should be, it is the
43:24
only weapon we have at the moment as
43:25
humanity for decentralization.
43:27
There is nothing else.
43:28
Yeah, it's it's very
43:30
interesting to me and I'm sorry cost
43:31
us I know we're probably eating into
43:33
maybe some other time this morning
43:36
but you know just the thought
43:38
of the the
43:41
The closed nature of these
43:42
businesses the Anthropics the open
43:43
AI is the deep seeks etc That
43:46
own and control the most powerful
43:47
models These models probably
43:49
wouldn't have been as powerful if
43:50
they as they are if they hadn't
43:52
had access to all the people using
43:54
Them right because the models learn
43:56
from the inputs And if we went
43:58
to a decentralized alternative, like
43:59
QVAC is kind of positing,
44:02
where the user keeps all of that
44:04
information and data on their
44:05
device, it never leaves their
44:07
device. The model itself never gains
44:08
access to that information.
44:10
Would it have grown
44:12
at the exponential rate, the
44:14
capabilities of the AI,
44:17
as opposed to how Anthropic and
44:19
OpenAI and DeepSeq and, you know,
44:21
at Kimmy.
44:22
All these frontier models have,
44:24
because so many people have been
44:25
using them. Because as my
44:26
understanding goes, it's
44:28
the inputs that really matter.
44:29
It's the sheer amount of people that
44:31
are using them that are helping
44:32
these AIs grow in
44:34
understanding and capabilities
44:36
so fast.
44:38
I was, Gareth, in this dilemma
44:40
for many months in the last
44:42
few couple of years, right?
44:43
I don't know, to tell you the truth,
44:45
what is better.
44:46
There is a problem.
44:48
AI is energy hungry.
44:50
All of this model, by the time
44:52
you're adding decentralization, of
44:53
course, as you know, I know it very
44:55
well, you have to optimize on
44:57
everything because you cannot easily
44:58
be in par with the current
45:00
non-decentralized architecture.
45:02
Everything that is decentralized is
45:04
more expensive.
45:05
And AI is already more expensive,
45:08
right? It requires so much access
45:10
to GPUs.
45:11
We are not controlling even,
45:14
like, you know, why do they have
45:16
limits on the context, right, and
45:18
the number of tokens?
45:19
Because you need the energy.
45:21
It's a cost, right.
45:22
It's cost. By the time you're
45:23
getting into decentralization, this
45:25
will be multiplied by something,
45:27
right depending on the, you now,
45:29
Nakamoto coefficient or whatever we
45:30
consider decentralization.
45:33
Then eventually you end up into a
45:34
situation where you have to balance.
45:37
Like how much time do you give to
45:39
these companies to run like
45:41
a proprietary software,
45:43
like closed source and
45:45
improve the energy consumption of
45:47
these models because they need to
45:49
optimize as well compared to going
45:51
directly to decentralized, yes,
45:53
protect the data of users.
45:54
But at the same time having, yes,
45:57
more data but slower
45:59
like reasoning just
46:01
because For decentralization, you
46:03
have to pay a penalty, right?
46:05
So we need, I
46:06
don't know when this bar will be
46:08
met, but there is
46:10
a race between the energy
46:12
consumption, what you can
46:14
do to audit these companies.
46:15
It's not always so easy because they
46:16
already have so much data that this
46:18
data might be so private that it can
46:20
be a weapon by itself.
46:22
So it has to have some control.
46:23
We don't want to give all of this
46:24
into hackers.
46:26
And at the same time, yes, it
46:28
can on user's device, but by the
46:30
time it's on the user's devices.
46:31
Then you rely on
46:34
the CPU and the GPU that it's in
46:36
some device that is very cheap,
46:38
right? It's $1,000 best
46:39
case scenario with iPhones.
46:41
Then eventually this will slow down
46:43
improvement and evolution into green
46:45
AI.
46:47
So it's a race.
46:48
I believe both of them will exist.
46:50
I want to see the decentralized
46:51
version of the full AI, not only
46:53
memory that we're doing here, like
46:55
whatever others are doing with
46:56
putting AI on user device, at least
46:58
the data. But I don't think by
47:00
itself and alone We're still
47:02
at the state that AI is
47:04
very hungry. Imagine that.
47:06
I don't remember how much do we get
47:08
as humans. Is it 20 milliwatts or
47:10
something?
47:10
The human brain to think of
47:12
something and burn energy for this.
47:15
AI is already hundreds
47:17
of times more expensive.
47:18
We're not there.
47:19
We're no faster than biochemical
47:22
reasoning, right?
47:23
So we need also to
47:26
invest into making these things
47:27
fast. Unfortunately sacrificing
47:29
and creating a grace period for
47:31
these companies to work.
47:32
At the same time, I cannot imagine
47:34
just opening a lane for
47:35
these company and not the open
47:37
source and the decentralized lane.
47:39
We need to do it on parallel.
47:40
But there will be, let's be honest,
47:42
there will for a while a period
47:44
where all of them
47:46
will progress on parallel and
47:49
eventually, I don't know, by
47:50
convergent evolution or by whatever
47:52
we believe is, as humanity is,
47:55
sane and wise, we're going to make a
47:57
pick one day.
47:58
This is happening with all of the
48:00
technologies. In my opinion, this
48:02
eventually, they end
48:04
up into a small
48:05
cycle of people who have this power.
48:08
But there is also a new
48:09
do-it-yourself evolution on
48:11
the site. Until the do-
48:12
it-yourselve, imagine Europe is
48:13
doing everything for green energy
48:15
and all of this stuff.
48:16
But you will never see a Google or
48:17
Facebook in Europe because this goes
48:19
against regulation.
48:21
Europe is very sensitive with Mika
48:22
and GDPR on this.
48:24
So we need, actually, I believe
48:26
that different countries or
48:27
continents will follow different
48:29
paths,
48:32
just because of how their society
48:34
is built.
48:34
From full capitalism to very
48:37
green energy,
48:39
focused and sensitive to, I don't
48:41
know, to socialism and a few other
48:42
things. Unfortunately, it will end
48:44
up there because AI will
48:46
affect society.
48:47
It's affecting society already.
48:49
We don't even know what will happen
48:50
with unemployment, right, where I
48:52
was talking, I'm at.
48:53
Like I'm talking with some like
48:55
French ministers in various
48:56
countries of course if you if
48:58
you know cryptography you can also
49:00
help like on protection
49:02
of like governmental
49:05
data infrastructures and so on.
49:07
Nobody knows what will happen.
49:09
Like we have this interview now
49:11
about the effect of
49:13
closed source decentralized open
49:15
source. Now we even have three
49:16
lanes.
49:17
You don't know because in different
49:19
jurisdiction different countries one
49:20
of them is more profitable or
49:22
a better suite, better fit for this
49:24
particular continent, but not for
49:26
the other.
49:27
So eventually, I
49:29
believe it will happen in all of the
49:30
parallels.
49:31
And then based on where you live or
49:33
where you want to live, this is what
49:35
you're going to do as human.
49:36
And we're going see innovations in
49:38
all of the three directions.
49:39
It's three directions, I'm telling
49:40
you. It's not easy.
49:42
Not an easy peak, not in this
49:43
transitional grace period
49:45
where we don't know exactly which
49:47
one is better.
49:49
Yeah, no, we don't.
49:50
We don't cost us and
49:52
I think we're gonna have to book you
49:54
on the show again in the next
49:56
few weeks or months to
49:58
continue this conversation.
49:59
Otherwise, we'll be talking for an
50:00
hour and a half or two hours.
50:02
I really appreciate you taking the
50:03
time to chat to us.
50:05
And yeah, I was really enjoying the
50:07
conversation and I'm loathe to end
50:08
it. But I know that I would take
50:09
another hour of your time.
50:10
So we will leave it there and
50:12
we will commit to having another
50:14
conversation sometime soon.
50:16
Yeah, and really appreciate
50:18
your insights and everything that
50:20
you and this select special
50:22
group of people are doing for both
50:23
the cryptocurrency and blockchain
50:25
industry and its convergence
50:27
with AI. Although I think AI is
50:29
converging with everything, which is
50:31
the crazy thing, to be honest, I'm
50:32
not even sure like I'm doing a
50:34
crypto show anymore because every
50:35
day I'm talking so much about AI.
50:37
Not even myself.
50:38
Am I doing just crypto?
50:39
You believe now?
50:40
You see, Walrus Memory is one of my
50:42
kids, and there is also quantum
50:43
protection and so on.
50:45
Next time, I'm going to discuss
50:46
something very, very interesting,
50:48
pretty much in the same lines of
50:50
conversation, why AI
50:52
is our ticket to open quantum.
50:54
But I leave it here as a
50:56
teaser, and I will explain why
50:58
humans cannot go to the quantum
50:59
revolution, but maybe humans and AI
51:01
can go.
51:03
Beautiful, definitely.
51:04
I'll bookmark that.
51:05
Thank you so much, Costas.
51:06
And thank you for joining us on the
51:08
starting block today.
51:09
We will be back next week.
51:13
But I do want to make a public
51:14
service announcement. My wife is
51:15
expecting a third baby anytime soon.
51:17
So it could be that we take a
51:19
two-week break from next week, but
51:20
you can expect us back on Monday.
51:22
If not, you'll know where exactly I
51:24
am. Thank you, so much for joining.
51:25
If you want to catch up on any of
51:27
the content or interviews, you can
51:28
find that on our YouTube page or
51:30
our ex account. To go ahead and
51:31
check that out. Like, follow,
51:34
retweet, subscribe on whatever
51:36
platform you're joining us on today.
51:38
We'll see you next week. Have a
51:39
fantastic weekend.