Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
AI Power Struggles, Quantum Leaps, and Web3’s Next Reckoning🎙️
Recorded: Aug. 19, 2025 Duration: 1:39:52Space Recording
Short Summary
The latest discussions in the crypto space reveal significant trends in AI development, including the launch of DeepSeek 3.1, the emergence of Claude as a competitor, and the growing importance of private data in AI training. As AI technologies evolve, the need for improved reliability and regulatory frameworks becomes increasingly critical.
Full Transcription
Thank you. Thank you. Thank you. Thank you. Music Thank you. Music Thank you. I'm going to go to the next video. Thank you. Thank you. Thank you. all right let's get started.
Lewis, how are you doing today?
Oh, nice.
We got David again.
Bring him up.
How's it going?
Good.
How has your experience with GPT-5 improved over the last week?
Mine has not, and I've now seen that they've added the 4.0 back.
Yeah, they've added it back so you can select it.
So it just seemed a little slower than 4 for me doing my usual work every day.
So now I'm just flipping back and forth when I know I just want it faster
and it's really going to do the same thing.
But 5's been interesting.
You know, the answers are better. It does seem to think better, but that's just, you know,
maybe that's just me. David, have you been having a chance to use it?
I've been playing around with it a little bit. I tend to lean in more on Brock and some of the open source models, just because I can run them
directly and have API access. But it sort of seems like there's been mixed reviews.
It's still sort of limited by their philosophy and the way that they, you know, curate the data
still sort of won't let me do a lot of the things I want to do with APIs or keys or engineering tasks.
So we'll see.
Yeah, I mean, for me, I use it to run some limited Python work, and I just haven't found
any of the other chatbots that do that.
So when I'm doing modeling, it's just easier.
I do want to mention that there was some breaking news this morning about four hours ago.
DeepSeek 3.1 was officially announced, which we all kind of predicted.
You know, once ChatGPD 5 came out, you knew DeepSeq was going to come out.
They've added more token memory.
They've got, you can produce, process more massive documents.
Apparently, it'll handle longer transcripts and data sources.
They've improved the memory handling and speed for AI workflows,
and they've enhanced the summarization to give some apparently
more accurate insights.
And you can scale research and analysis like never before.
So that's now out.
So, you know, these power plays back and forth between, you know,
ChatGPT and DeepSeq and Grokk and everyone else it's kind of interesting watching
them juggle back and forth uh what do you think noah i so with gpt5 i found the hallucinations
to happen a lot more frequently and it was getting yeah i mean and maybe it's because I'm, I mean, I'm, I'm asking it to help me with very specific things, but it just kept getting stuff wrong to the point where I was getting frustrated and I didn't stop using it.
Um, but I just did research without, like, I just, just got off the AI and, um, did research without it.
And then I kind of copy pasted the correct answer.
It learns like it,
if you correct it,
it's going to remember that and it won't make the same mistake.
But I just found that the number of mistakes have been a lot more than with
4.0.
I noticed last night that 4.0 was back because I was,
I was trying to set up an LLC and I was using it to help me with that.
But otherwise, I don't, I mean, I know Ryan speaks highly of Grok.
I don't really use Grok that often because anytime I do, I just, I don't find it to be as efficient as GPT.
And then I've tried using Venice AI as well, which is great.
I just, it's not as powerful as GPT either, at least in my experience.
Maybe David or Lewis, you guys have experienced something different.
Well, it feels like maybe the only real rival it has is Claude.
Claude, interesting, interesting.
I mean, you kind of bring up an interesting point point noah where we kind of talked about this before that if you're doing tasks on ai and you have the domain knowledge
skills and experience you know when it's hallucinating you know and and the big concern
is you know people using chat gpt or really any chatbot and and trying to get answers uh
and trying to get answers for questions in which they're not experts.
I mean, the hallucinations are sometimes super obvious,
but I'm sure there are cases where they're not very obvious.
We saw the person who said, give me a healthier alternative to salt,
and they ended up going into hospital and other weird stuff happening,
and kids asking you questions, you know,
when they're going through a life crisis and then they're getting the wrong
answers. You know, it brings up, you know,
how do you regulate and how do you kind of do this? But, you know, Noah,
do you think, do you think we will ever be able to kind of regulate this?
regulate this or do you think just the AIs will get better so they won't give us such crappy answers?
Or do you think just the AIs will get better so they won't give us such crappy
answers?
I think it's a loaded question, right? Because will we be able to, yeah, so I think it will get
better where you can trust the answers far more than you do right now. I mean, I think that's
inevitable. In terms of like being able to control it, I don't really, I don't know. I think that's inevitable. In terms of being able to control it, I don't really...
I don't know. I think I've watched too many science fiction movies and I don't know enough
about the subject. Or rather, I don't know enough about AI from a developer's perspective,
because I'm not a developer, to be able to predict whether it becomes sentient or whether it becomes self-learning and self-improving or whether it starts to view humans as inferior.
I mean, all these different speculations are just that in my mind.
I don't know what direction we're heading in, but I can say with almost certainty that I do think things like hallucinations, the more difficult the subject,
I think you'll still have them.
But in terms of, I don't know,
completing a task or creating an LLC, for example,
I think in the future,
you'll just be able to say,
make this LLC for me and it'll do everything it'll
do everything with a click as opposed to you having to go step by step which i do prefer i
mean i like i like doing stuff like that it's um it's stimulating for me in a weird way i love
setting up llcs and companies for my friends and family at the time, it's nice to have this AI agent that streamlines the process.
Right.
Well, I mean, that's where you're dealing
with a really well-defined problem.
You know, when it's scoped out, it's repeatable.
You know, people have done it thousands
and thousands of times before.
You know, obviously the hallucination piece
is where you're dealing with, you know,
what I've talked about, the unknown unknowns,
especially in respect to your own context. One thing I do want to bring up, there's a thing in
the headline news today where Sam Altman was talking about that AI is going to bring us more
abundance, time, and bigger families, which is kind of interesting. But I mean, I'm old enough
to remember the mainframe days and then, you know, in the 80s.
Remember when PCs came out in the 80s and everyone said, oh, these PCs are going to save us time and we'll be able to go on holiday more and all that.
And it seems like every technological revolution we get, you know, the proponents always say it's going to save us so much time and we're going to have all this extra leisure and it's all going to be wonderful.
time and we're going to have all this extra leisure and it's all going to be wonderful.
And I'm just wondering if we're in a rinse and repeat cycle here with AI that, you know,
while they're saying it's a, you know, it's a liberating force and it's great and, you know,
all this sort of stuff, essentially the structural incentives haven't really changed,
you know, and history tends to tell us that, you know, it's just going to get absorbed into the
same profit maximizing systems that have really resisted reducing work hours for decades. I mean,
do you honestly think that we're going to have more leisure time? You know, that we're not,
you know, things won't shift and then all of a sudden we'll just spend day, all our day,
just managing AI bots. I mean, what's your thought? So as an economics nerd, there's a great term
for this. It's called Javin's paradox, right? Where the cost of something goes down, let's say
the price of compute, right? Computers got faster and faster, but the absolute amount of money spent
on them actually increased, right? Because as compute got cheaper, great,
now we can do high quality videos or whatever the new application was, right? So people have
actually increased their spending every time computers become cheaper. And so there's a couple
of points here. From an engineer's perspective, let's say the average engineer could produce a
quarter million dollars worth of software each year, and he was getting paid 100, 150K, right? Now, if he's five times as efficient,
he can produce more than a million dollars of software per year. Do you want more engineers
or less engineers? You want more engineers. So we're going through this dip, right? Where there's
a bunch of engineers that haven't been trained on the AI tools yet, but they
are rapidly adopting those tools.
And the ones that do become massively more efficient and valuable, right?
To the point, it also reduces the barrier, right?
When we got Excel sheets, all of a sudden you didn't have to be an accounting expert,
right?
To maintain a basic spreadsheet, right? Okay, great.
Now we all get to be coders because all of a sudden you don't need to write the
code by hand.
You just have to be able to articulate what you want to create.
And so I was playing with a great new tool that came out recently called
GenSpark AI, and it's basically a Claude wrapper,
but they've put it in this great format
where it's me using it with my data and I can connect in the Morpheus API and I can power it
all with, you know, Llama 3 or whatever model I want. And I was walking my nephew through how
to build a website. And within two hours, we had not only built the website,
it integrated all the hosting, the database,
and then we just handed it the key for Morpheus
and it hooked it up to the API.
And he built an AI tutor for his fellow students
in a matter of a couple of hours and it was live, right?
And that's the first time I've seen it going all the way
from not just the code, but be able to host it,
be able to actually connect it to an API.
And he's not an engineer, right?
He doesn't have a technical background, right?
But just a little guidance, you know,
push this button, copy and paste that, done.
You know, now all of a sudden he's teaching all his friends
how to use these tools.
So personally, I think it's gonna be amazing, you know and it's going to add to a lot of value to people. The last point would be, okay, yes, we have not reduced our work hours. In fact, we've increased our work hours because, okay, now I can earn $100 an hour or $200 an hour instead of 10 or 20 like in the past.
10 or 20 like in the past, right?
So it's a personal incentive.
It's not just, you know, sort of the corporate machine.
Like how much money do you want to make?
How much time do you want to spend on vacation?
And, you know, Americans and Europeans have come up with very different answers.
But if it's my AI, my personal AI that's working on my behalf 24-7, 365, it's like I'm continuing to work as long as the value flows back to me,
right? That's the magic part about these. And so, yeah, I might spend a bunch of hours fine
tuning and telling it what I want to accomplish and, you know, hooking it up with more tasks and
stuff like that. But it may well be the exception to the rule we've seen before, because yes,
I'm putting in more hours, but it's not my brain power or my physical energy. It's my agent acting
on my behalf. At least I think that's the golden path we really want to head down.
Right. So it seems like the need for certain for work to do things will always be there i mean i it's like you know when you've
seen uh cities you know they they have a two-lane highway and then they add a three-lane highway and
a four-lane highway and it just fills up i mean the traffic doesn't get less and you know it's
it's it's you know so what you're saying is while these tools will increase our output and they make us more efficient, the workload's just going to rise.
Go ahead, Dave.
Sure.
And, you know, I'm okay with that as long as the value is flowing to me.
Right.
And that's why I think having personal ownership of your AI is like the thing.
Right.
Either you're going to own AI or AI is going to own you. It's going to
go one way or the other, right? And so I'd really like to have a personal agent where I can store
my outcome, my knowledge, my tasks, everything that I want to accomplish and then have its
outcomes be aligned to me. I put out on X the other day that we should define AGI as artificially generated income, right? I define AGI as when my AI
produces more than half of my income, right? That's a really interesting stepping point because
it's very practical as opposed to like measuring how smart it is. Is it smarter than most humans
and most things? Yeah, probably. Well, if that's true, then it should be able to do more than half my job.
And so I'm going to test this out next month.
I'm going to start a hundred day race to AGI starting on September 2nd in honor of the
two year anniversary of the Morpheus white paper.
And I'm going to see every day if I can automate a task with my agent and if I can get there
before the end of the year
and basically have handed off more than half of my workload that I earn my income from to my agent.
So it's going to be an interesting experiment.
Right. Now, that brings up the whole thing about e-commerce and agents and, you know,
how do you set up the rails so that, like you said,
you've created your AI, you're doing your AI work.
How do you ensure or what's the process and mechanism to set up
so that the e-commerce flows to you?
I know Walmart's obviously working on this.
There's lots of stuff in the news about be prepared
that most of your e-commerce is going to start being done by AI agents.
What are you doing to set up so that as your agents or as this AI works,
that, you know, there's a little, you know, PayPal thing that pops up or whatever it is?
Well, I think the key is you have to own the wallet, right?
So if I have a smart agent,
right, an AI that I've connected to my crypto wallet, to my smart contracts,
then it has a way to pay me and also for me to be able to pay other agents, right? So now you have this agent economy going. But yeah, if you don't have a wallet, then you're not participating
in that economy and you're just being farmed, right, for your data, for your information, you're paying for compute, right?
You're slashing your credit card on ChatGPT and paying them 100 bucks a month or whatever, right?
So I think that it goes back to do you have, you know, self-sovereignty?
And to express that in reality, that means having a Web3 wallet, right? And if you've
got that, which now 700 million people do, right? The adoption is really accelerating.
And this is just going to make it easier, right? If an AI can help onboard people that are less
technical, that's sort of like that magic search engine moment was for the web, right? People were
dismissive of the internet. They're like, oh, it's just for engineers. It's just for university kids. Nobody's ever going to use it.
Then search engine showed up and that made it accessible to billions of people. So I think
we're at that same sort of inflection point. So hopefully we're seeing this convergence of trends,
right? Where the AI is natively using all of these smart contracts and other Web3 tools,
and it's onboarding more humans to have that type of autonomy.
So I think that's the solution.
Sure. Captain Levi, go ahead.
I think he rugged.
Is he there?
No, no, he rugged. I agree with's, I think he rugged. Is he there? No, no, he rugged.
I agree with you, David.
The thing is, you know, the thing that made sort of commerce at the retail level easy was we had MasterCard and Visa, right?
Very simple.
I got a card.
I flash it.
I'm done.
I'm out.
I don't.
Yes, it's true.
People have wallets.
Yes, absolutely true. people have wallets. Yes, absolutely true people have wallets,
but I don't see a sort of a universal e-commerce mechanism
that allows everyone to kind of easily do transactions via wallets.
I'm not seeing that yet, and I think unless you have,
or unless Morpheus is going to design that, give that to us.
But I really haven't seen any AI e-commerce wallet integration yet that would allow me to go, oh, I want to buy this from you, David.
Let my wallet talk to your wallet and bang, we're done.
I mean, what do you think?
It's definitely getting there. So I would highlight Coinbase's recent work on what's
called the 402 standard, right? So the early days of the internet protocols, there was this idea
of reserving a code, just like you have 404 for an error on a website, to reserve a code for payments online.
And that code was 402. Now, back in 2000s, early, late 90s, there just wasn't Bitcoin. There wasn't
the infrastructure. So it kind of sat there. Nobody used the standard. But now Coinbase is like, hey,
guys, we have all the pieces now. Let's do internet payments with just a link, right? And we'll
power it with stable coins and you'll use the 402 protocol to do payments. So they've connected that
to AI agents, right? And a lot of people are racing to integrate 402 as a way for their agents
to handle those payments. So, you know, I think all the pieces are there and coinbase is probably
in the best position them or stripe of any of the mainstream companies uh to onboard people
uh in mass to using that and they may be using that without even knowing they're using that
right if they've already got you know a credit card stored in their their computer or their
browser you know uh it's just a matter of handing credentials through
Stripe or another provider. And so once you have that infrastructure, you're going to see
agentic payments really take off. But the point I'd make on the decentralized side
when it comes to Morpheus is I think about Morpheus as protecting people from extractive AI, right? Because if you have a personal AI
watching out for you that knows what you're trying to accomplish and is aligned to you,
right? It's a fiduciary to you, right? Its highest outcome is to serve you, not Google or some
company. Then you can really have commerce takes off. Just the way in the internet, people really didn't take it
seriously until PayPal, right? PayPal showed up. They're like, hey, you can safely use your credit
card here. We'll make sure if the merchant doesn't send you the money, you get your money back.
And for the merchants, right, they were going to handle all the payments and the complexity.
the payments and the complexity. And so e-commerce and online was really born out of that type of
risk reduction that PayPal provided, right, to everybody involved in e-commerce. And so we
definitely need a similar set of infrastructure. And I think that's what's coming together.
Okay. Okay. No, I get it. You know, that's, I think from looking at it, you've got the HTTP 402 that gives you your web standard, you know, payment.
But then what I'm seeing is Coinbase is connecting it with the EIP 3009.
And that's the blockchain mechanism.
So it's, you know, how do you authorize?
So it seems like it's the combination of both the 402 and the 3009 that's going to give us this integration with AI agents to be able to do this commerce,
like you're saying. Would that be right? Yep. That's exactly it. So, you know, they've been
making a bunch of announcements. And you think about how far we've come. You know, Bitcoin
block times are 10 minutes, right? The original Ethereum block time was 12 seconds. And now you've got lightning fast transactions on base that are a fifth of a second. And so we're getting pretty close to instant transactions on blockchain as far as human timelines are concerned. A fifth of a second basically seems like instantly it happens.
you know, a fifth of a second basically seems like instantly it happens.
Right.
So, yeah, that's the combination that really makes this interesting.
Right.
Authorization from the agent using, you know, the Ethereum standards you were talking about
and on the other side, integrating it with Web2.
Right.
Yeah, I think, you know, if Coinvoys can do this and produce, you know, something that
we can then just copy and
paste and integrate you know it's just going to make life a lot easier um or maybe someone on
this call might think of it and actually do it for us which would be great but you know i think
until we get some sort of universal or accepted like you know we were talking about you know as
as new ai agents come out you know how do you trust them or not?
There's the brand, right?
People trust somewhat the brand of MasterCard or Visa in finance, right?
Because we know they've spent billions on security and they've got all the network.
We might need something similar to that for this combination of the 402 and 3009 to enable, you know, acceptable and
integratable AI agent commerce at some point, which will be really super interesting.
So, yeah, I think you're right.
I think, you know, we've talked about this whole thing about, you know, AI is going to
give us, you know, some free time and leisure.
But I think at the end of the day, once we get through this really sort of seismic shift,
we probably will keep working, you know,
40 hours a week or whatever it is,
but hopefully there'll be more choice around it
than anything else.
Let's see.
One of the other things-
I hope not.
I mean, I think if AI isn't able to give human beings a much higher floor in terms of quality of life and doesn't give human beings, again, a higher floor of choosing what to do with their time, I don't know. I just, if, if human beings are still working 40 hours a week
in, um, I don't know, like honestly, even 10, 10 to 20 years, cause the stuff's supposed to
evolve rapidly. I think we've kind of, I don't want to, I don't want to make an absolute statement,
but I think it's kind of, I think it's kind of an L like it's an L to me.
Yeah. Yeah. I mean, I did a LinkedIn post post the other day you know where you know there was this guy like windsurf you know the new ai coding company that got bought i know billions and and the ceo
um whose name will not be mentioned um said you know he wanted his people to work 80 hours or 90
hours a week and i just went that's just completely bogus, right?
I mean, there's so many studies now across so many countries where they've actually reduced it.
They've tested on four-day working weeks,
and they've found that people are more creative, they're happier,
and they actually produce more revenue.
And so it's just weird to me that we have these 18th century mindsets
that, oh, you know, you're my slave, you know,
I'm just going to work you to the bone and I'll toss you aside when you're exhausted.
And it's like, no, that's not effective anymore. You know, we have so many great tools. We have
so many things we can use now that when people go to work, they can be incredibly efficient and
they can certainly do it in four days and they can certainly make you tons more money. I mean, you know, what's your thought? Well, have you ever seen the video
by CPG Gray about weekend Wednesdays? No. I highly recommend that video because it's talking
about what he just mentioned, which is people are more productive if you break up the working week.
And he wasn't necessarily proposing a third weekend day on Wednesdays, but he was basically
saying if you shifted it for Saturday, you'd get all these benefits. But it's not a bad idea
to kind of claim that meme, weekend Wednesdays brought to you by AI, right. And have that as an aspiration,
right. You know, people used to have almost no days off and then, you know, everybody got
universally Saturday or rather Sunday off. And then only recently in Japan and Korea,
a lot of these societies have only recently embraced the two day weekend. It would be
interesting to meme and sort of connect that in everybody's mind that the benefit
of my AI is I'm taking Wednesday off, right? I'm going to go in Monday and Tuesday and jam and get
some stuff done and the AI can take it on Wednesday and I'll be back on Thursday to Friday to check in
and then I've got my weekend. That would be a really interesting, you know, societal improvement, you know.
And so that's a great idea.
I would love to meme that.
I think that would be extremely valuable.
Weekend Wednesday.
I'd prefer an AI Friday, to be honest.
But, yeah, I just was amazed that this that this you know CEO came out with this statement
in this you know that oh I want them to work 89 hours of nine hours a week and you have to do that
and it's just not true it's just absolutely not true and I just think it just dehumanizes the
people that you work with and I it just seems to be this, you know, abusive sort of
mindset that some people have that it's like, I'm just going to, you're just a tool to use,
and then I'll cast you aside. And I think today, in today's economy, in today's business, you know,
you have to come from a level of compassion, and and support and you get the best out of people
and like I said you end up making more money I you know so anyway that's my little soapbox rant
there was some stuff that came out over the last seven days around quantum computing and some
really interesting breakthroughs with quantum accelerated models that they drastically reduce
training times now you know obviously we're still
in early days and all that sort of stuff with quantum computing but um you know they're saying
quantum's going to scale soon they expect it to start you know you'll be able to scale up on the
current chipsets they've got um but you know when you start to do your models on quantum computing and you know they're the training times are
shorter you can scale up the wazoo with those um no do you think you think the shift from you know
the infrastructure of our current you know uh turing machine um pc mindsets you know single
flow sort of stuff to a quantum, quantum accelerated computer is
going to make a real big difference to our models. I don't know enough about it, to be honest with
you, Lewis. I don't know enough about quantum computing. I haven't read, I'm not an engineer.
So if you recommend any good books that aren't, or any good literature that isn't oriented towards engineers
that would be great but conceptually I just don't know enough about quantum computing I understand
people say it's going to be able to crack different encryption or different kinds of
encryption and people are worried about old bitcoin wallet addresses but I just don't really, I want to know more about it.
Go ahead, David. Any quantum in your life?
Yeah, as the engineer nerd on the call, I'll be happy to speak to that.
Cool.
So, you know, when it comes to cracking, I'm not too worried about that. So NIST, which does the National Institute for Standards in the US has already released all
their recommendations on post-quantum encryption. And the military and other people have been
implementing that since like 2018, 2019. So most of the sort of serious players and big tech
companies are moving over to that encryption. If you're using Signal, private messenger, for
example, you're already using post-quantum encryption. They did their upgrade, I think
it was last year, so that they're all ready to go. There's a proposal, Bitcoin improvement proposal
for them to adopt post-quantum encryption. That's a lot more complicated, not for the wallet
addresses, but for the miners that will require sort of a big change.
And that's a bit further out, but at least securing the wallets and the encryption on Bitcoin,
you know, there's already proposals for doing that on Bitcoin and Ethereum.
And so when it comes to AI, I think certainly increases in these efficiencies in the models
will be extremely valuable,
but we're sort of running out of public data to train them on. There's about 10 trillion tokens,
you know, language tokens on the internet, right? And the big companies, they've sucked in
all 10 trillion tokens and trained these models now on effectively the entire
corpus of human knowledge, all the books,
all the internet, all the videos and everything else. And so what's next has a lot more to do
with private data. The amount of private data dwarfs the available public data, right? The
information you have on your computer, in your personal cloud,
enterprises have in their servers is much larger than what's available on the internet. And so
if we can get sort of fully flushed out decentralized AI, where people are confident
that this is their data, they're not giving it up, they're not losing access to it, you know, by
connecting into an AI, we're going to unleash effectively this next wave. It'll have a lot
more to do with personally benefiting the person that holds the data than going out there and
scraping the internet. Well, you're right that they're running out of data, but I mean, we're creating data, new data every moment going on.
Like someone said in 2023, there was 120 zettabytes of data was generated globally.
So you're right that they might have run out of existing data, but I think the continually created data, both by humans and by just machine data, which is three times larger.
Surely that's going to keep, you know, every, all the AI busy, you know, running along merrily. No?
Oh, yeah. You know, the amount of information on the internet, you know, I want to remember the
statistic, I think, was doubling every year or two, right? So, you know, the amount of information
is just an incredible,
I mean, that's why they're building
so many data centers.
I don't know if you saw that,
but data center construction
has surpassed capital expenditure
of all office workspace
in the United States, right?
We are spending more money
housing the computers
than now we are the humans.
And that's sort of indicative
of where we're
going. So yeah, you're right. It'll go from 10 trillion tokens and next year it'll be 20 trillion
tokens and a few years it'll be 40 trillion tokens. But the fact remains, the private data
is 10X that, right? 100X that. Just think about yourself personally, like how many photos you
posted to Twitter versus how many you have saved in your iCloud?
I've maybe posted a thousand pictures publicly on Twitter the last 10 years.
I've got a hundred thousand images saved, you know, in my iCloud since, you know, 2007 or something like that.
Right.
So, you know, that's going to be an amazing wave because what you're going to get is personalization, right? That's the big unlock, you know, these generalized models. Okay. They know about the world in general, but I want one that knows me, right? Knows me even better than myself. You know, my preferences, where I've been, you know, who I know. I think of a practical example. I go to a conference and I speak at five or 10 a year. I've spoken at 100
conferences the last 10 years. I've met something like 10,000 people over that period. I just look
back at the contacts in my phone or my Rolodex on LinkedIn or X. It's like 10,000 people.
And there's no way that I could keep track of all 10,000 of them. But when I'm building a new
project or working on something, I would
love to remember that, oh, I had a conversation two years ago with a cybersecurity guy who knew
exactly this subject matter. I should reach out to him or I should send him this paper.
And so I think that's going to be an enormous unlock is when we have our personal data
securely with our own AI working on our behalf. That's
just going to be amazing. I use agents a few times a day today. I think by the end of the year,
I'll be using agents hundreds of times a day. And I wouldn't be surprised if it's thousands
or millions of times on my behalf in the years to come. So that's what gets me excited.
So that's what gets me excited.
Where do you...
Go ahead, Lewis.
No, go ahead, Noah.
I was going to say, David, what's your estimated timeline for that vision to come to fruition?
Well, like I said, I think by the end of the year, the majority of the things I used to do in my day-to-day work will be done by my agent,
right? I've already got an agent. Can you give some examples, please?
Sure, sure. Like, so there's great tools like Zapier, take for example, right? And I can use
Zapier to spin up an agent, and I have, that reads my email every day. And instead of me reading,
you know, through all the junk mail and
notices and newsletters, it gives me a daily summary of my inbox. And if I want it to draft
or apply, you know, I have an agent that can then draft or apply. You know, I remember organizing
an event and I needed to send out a message to 400 people that I knew in the industry.
And normally that would have taken me hours and I would have set it to template up,
template up on, you know, some sort of mail service.
Instead, I just created an agent and I hooked it up to my Gmail and I let it send the message directly from me, from my inbox, right?
Not some lame template on a newsletter, you know, kind of thing, but a message directly for me.
And the open rate was sky high compared to like if I had used some tool to send out that same message because it was from me.
Right. And I got that done in 30 minutes instead of, you know, three hours of work that would have would have taken before. So these are just simple practical things where I'm connecting my apps,
which contain all of my data to the intent and the knowledge of a large language
model. And here's what I want to do.
Right. And boom, it can do it.
So I'm going to make a concerted effort this fall to basically break down a lot
of my regular work. I'm in meetings.
I've got an agent taking notes.
Afterwards, I have an agent reading the notes and putting it into a report that gets posted to,
you know, Slack or Discord or whatever. David, you, I mean, you seem, you seem to
obviously be ahead of the curve and have integrated agents into different kind of different decision making roles within your day to day life for people listening or for people that don't know how to get started.
Because it's one thing to use agents for development.
But the example I think that you just gave with the email is incredible because so many people can relate to that, right?
I have to go through my email and I somehow end up on junk lists.
I'm actually frustrated about my Mobi Media email because the only time I use that email is for crypto conferences or events.
And I feel like one of these events just sold my email to random bots that just email me all day.
How can someone like me or someone listening get different ideas or just find tutorials for integrating agents into day-to-day tasks?
So, you know, there's a ton of good material. I would point to community members like Luke Stokes
in the Morpheus ecosystem that's put together all these quick two, three-minute videos about how to
practically use these tools, right? engineering background with no technical knowledge required right um and just just be able to hear
push this button and now you've got an agent now you can do this this next thing you know so the
community in decentralized ai you know is is very helpful uh has produced all kinds of tutorials. Some of these are available on more.org.
It's just mor.org.
You can see what the community has built.
If you go to the builder section,
there's 250 projects now building on Morpheus.
A lot of these are agents.
A lot of these are tools that people are open sourcing,
that they want people to come and use.
There's an amazing amount of activity
going on. If you're going to conferences, a lot of them have a decentralized AI day.
So like consensus or permissionless or token 2049, all last year, everybody had a side event
around decentralized AI. So if you want to go to something in person, if you want to see a tour
online, all those resources are there. And I agree,
like the email example is very powerful because we all have to deal with it. Right. And it's a
proxy for our task list. Right. Somebody emails us, you know, your boss emails you, they need X,
Y, Z. Right. And you got to dig that out of the newsletters and all the other distractions.
So, you know, I would say, you know, teach yourself about this,
and I would prioritize it. Like, it's going to be the difference for how valuable you are
in your job and in what you're doing. The next few years, like, not having AI knowledge would
be like not having a smartphone or not having a laptop. Like those people will exist. They just won't be economically relevant.
I agree.
No, I certainly agree.
And tell my girlfriend this all the time.
It's going to be a gap.
And there's already a gap,
but there's going to be a gap between people that it's kind of the same with computers, right?
If you kept up with technology,
then you're better off today
than someone who hasn't.
Older friends that don't
really, they don't
understand this stuff the way that
a lot of my peers on
here, on Twitter, in crypto
do. So, I
want to
go to Captain Levi. I have some more thoughts,
but I don't want to hog them.
Like, Captain Levi.
All right.
Sorry.
I was on server and the call came in and I had to switch to Wi-Fi.
GM, GM, everyone.
So I'm just going to tailor on what Noah said,
because they usually said that knowing the problem is already half the solution.
So if I understand Noah correctly, Noah will probably need a really cool agent that goes through all of his emails and said that, yeah.
Let me set you up. Let me set you up for this.
Let me set you up. Let me set you up for this. I'm basically like as someone that is, I find myself to be inquisitive, but I don't know enough.
Like I don't know as much as David does where I can just speak, okay, this is what my day looks like.
And these are the agents and different AI tools I can pull in to streamline a lot of my workflow. Like I'm still not doing that. And so
the email example was one thing, but I, I genuinely, I generally, I generally think there's a gap
between people that want to integrate this into their lives and make their lives easier.
Uh, but don't know how to, I think there's a huge gap. Uh, and then people that do know how,
don't, do know how to, because I, I use GPT daily to complete tasks, tasks that you used to take, like preparing for interviews, right?
I've done like 2,000 video interviews in the last, or sorry, not video interviews.
I've done 2,000 interviews in the last like four years.
I used to take two, three hours to prepare for an interview.
Now it takes me like, I don't know, maximum 30 minutes.
And so, but that's know, maximum 30 minutes.
And so, but that's something I'm actively doing.
And I'm sure there are areas in my life where I don't even know that AI can make things
tremendously lighter.
But, and because of that, I'm trying to find out how I can fill that gap or how I can catch
up to guys like David or Lewis.
Oh, that. Okay. So basically, I think one thing that's going to help, like, if I understand the problem correctly by knowing setting up what to find saying
that these kinds of emails I am not really interested in them you know the agents would
probably you know go through the 200,000 emails that you receive and then you know classify
them that the last 15 15 000 emails about this person
uh talks about this uh or the last 20 000 emails about this person talks about this
you know um by carefully you know isolating it but then what you're saying i also detect a possible
ease of use because um charging between is so easy because all you need to do is just log in and have a conversation
and you're good to go. So I think the problem here to solve is that ease of use because
no matter how easy it is, I feel using these agents, someone needs to have some basic requirements,
some basic understanding. And one of those is how to use and manipulate apis and know what apis to use
um my most recent one was um talking about mcps you know uh model context protocols which uh
with the way things are going uh changes the entire concept of you know agenting AI so you know I feel of
course to some knowing the technical know-how of even a login or logging into
an email it's technical expertise enough it's a technical requirement so to speak
and building is agenting AI as I think are just some concepts that you know
people need to get used to
so that they can build and deploy, so that when you present the problem to these agents,
it knows the possible solution. But I'm pretty sure that something like that is in the works.
So that's my sense on it.
Yeah, Bo, go ahead. Oh, actually, sorry, I thought, no, Dirty, I think, was next. Go, Dirty.
Hey, Moby. I posted an article in there kind of explaining some of the differences of concepts of AI, whether it's super intelligence or AGI.
there's a blog article there
quite a difference there
with the LLMs that are
being used now
and the potential for those
and then what actually super intelligence
or AGI
the potential that they
possess so if you get a chance
just read that there's some links in there
a lot of good information
there conceptual stuff as well.
Awesome. Thank you. Thank you so much. Bo, go ahead.
Yeah. Hey guys. So this is actually really topical for me because the reason I was late
today is I was conducting an interview for my podcast with a gentleman who's founding a university
that is focused on AI first. And a big crux, and I don't say that to plug my podcast, although if you want to listen,
please do.
Plug it.
Plug it.
Plug it away.
Yeah, I promise there's a tie-in.
The reason I mentioned this is because a big chunk of our conversation was what it means
to be educated.
And that has meant different things over time.
So for example, prior to writing things down,
being educated meant you were able to regurgitate stories. You could remember stories and you're an
educated person if you can storytell. Then it became writing things down. Then it became
interpreting things other people had written down. Then we get into the invention of the computer
and it fast tracks our ability to do those things. And prior to consumer products, right,
prior to the Windows 95 moments,
the only people who are really using them
were people with some type of development
or programming skill.
But then we evolved into having operating systems
and front ends that are very user-friendly.
And now all of a sudden you go to college
and every single student is sitting in a classroom
with a MacBook or a PC,
if you're into that type of thing.
And so we're seeing infrastructure being built live.
Yes.
And so what I'm saying is you are not an educated person in the year 2025 for knowing how to use
a computer.
It is an expectation that you know how to use a computer to function in society.
And so what we're seeing today with agentic AI
is we're seeing the 70s, 80s, and 90s of computer development
where most of the people who are interacting with this stuff
at a fundamental level are developers
or are people who are leveraging these tools to build things.
And we have not yet seen a moment where the expectation
is that you must know how to navigate these
systems in order to be a functional member of society, or even not even that, but like a
somebody who, you know, thrives, let's say. But that is precisely why the super agent vision
exists. Because when that exists, that becomes the operating system moment for the end user, where I now have the ability to interact with these things.
And it learns my intent, and it helps me execute all the things I want to execute on without necessarily having to know the back end of how it functions.
And we're just not there yet.
But we can see it coming, I mean, without a doubt.
coming. I mean, without a doubt. Right. I mean, we're watching, in a sense, we're watching
Right.
consumer infrastructure being built live, and it's not there yet. We see the bricks, which are called
AI agents, the windows, I don't want to say that, the doors getting put in place, you know,
they're putting up, you know, sidewalls, all this other stuff. We're seeing it built live,
and we're all going right
when's this house going to get finished right because i really need it now you know right the
difference is with the internet there was no internet to show it to you so how were you going
to see it right right well the only mass media you know we saw ads for pcs and we thought oh i want
one of those but you're right. Now we have instant access to
all information globally. So now this level of, you know, and we're so used to instant gratification
that, you know, Noah's saying, hey, you know, why can I have this now? Where are these things? I
want them right now. But we're seeing them built and, you know, with AI agents and Web3 projects,
We're seeing them built with AI agents and Web3 projects.
We're seeing Web3 projects reinvent themselves with AI infrastructure.
I mean, Morpheus is classic.
It sounds like it's classic for that.
But we're also seeing token-driven compute marketplaces emerging
and a whole bunch of other stuff starting to accelerate
as things kind of merge together.
David and Bo, are you seeing, while you guys are kind of at that
forefront of Web3 and decentralized architecture for AI, are you also seeing sort of in the general
Web3 area that they're just kind of glomming onto AI and integrating it as fast as possible?
Well, you know, that's the great thing about open source is it's all built in public since day one.
Right. So, yeah, we do have a really amazingly transparent view of where the technology is at.
And yeah, you know, because AI and agents specifically are going to take over so much of the transaction volume.
Right. And we've seen this before. You know, who does the trades on wall street? You know, if you watched trading places back in the eighties,
you think there's a room full of guys, you know, bidding on the price of soybeans. Like
that room is empty now. That room is now a bunch of computer servers and it's almost entirely bots
trading stocks, commodities, everything, right?
So we've already seen this play out in other places and in other markets.
So it's not a big leap to say, hey, as soon as I can hand my agent, you know, a wallet
and it can go book my airplane, travel to the next conference.
And I don't have to sit there for an hour and look at all the
options. Like it knows my preference and it just does it. Boom. I'll do that the next day. Right.
And so people see this coming. They're like, yeah, you know, I'm a hundred percent going to do that.
So if you take that as the thesis, 99% of transactions, commerce, everything is going
to be agentic in, let's say the next three to five
years. Great. Well, if your project has anything to do with money or commerce or transactions,
you have to be relevant to where that's going. Otherwise you're running a dial-up business,
right? Which was funny to see. I think it was AT&T finally disconnected the last few hundred thousand people on dial service.
Yeah, dial is officially dead.
You know, they played the little ringtone and everything and posted it online.
I guess there were still a few people using it.
But yeah, like that's that's the reality is if you want to be relevant, you have to integrate these new technologies.
Sure. Sure. I mean, yeah, go ahead.
Dirty, I think, was next.
Hey, I just want to bring up a couple points.
So when we're speaking about LLMs, large language models that resolve problems through, you know, by what's programmed, right?
And determine what the answer will be.
And so all that being built right now,
the majority of what you're seeing is all centralized.
They're owned by corporations.
They're centralized.
They determine what the answer is going to be.
So looking into the future with general intelligence,
artificial general intelligence, is integrating the ability to problem solve itself, right?
And then also to consider that to decentralize that is very important, is that the information
that's being determined, right, or being figured out, the problem is being solved,
to be decentralized, isn't that the entire intent behind DeFi right now anyway?
And we're not really seeing it right now. Everything's being controlled centrally.
And so you have to consider that how it's being built and what is being built as well.
Do you want um to centralize
ai or not so consider that in your um as you move forward and invest into ai
yeah yeah thank you go ahead bud
yeah i would say the response to that, the difference between DAI development and DeFi development is DeFi is taking an existing legacy system that's been established for, let's call it 110 years or so, and then saying, hey, huge institutions that have had central control of money and the flow of money, you now need to adopt this system that's decentralized.
Whereas DAI is building alongside decentralized counterparts.
We're actually building the decentralized version and in some ways faster than the centralized
counterparts are able to ship it.
And so we're not trying to take over something that already exists.
We're trying to divert the way
that it will exist. And so I do think that's an important delineation to draw when we're comparing
it to DeFi. But by that same point, though, we are seeing DeFi adoption come into place. I mean,
Wall Street is figuring it out. And they will continue to do so. And eventually, you know,
everyone will figure out that fiat dollars are hot garbage and don't really mean anything.
And then we'll see the all at once moment, you know, the slowly at first and all at once.
Right.
We're also seeing in the news headlines last seven days, there's a lot of things coming up around AI infrastructure power struggles.
There's a lot of things coming up around AI infrastructure power struggles.
Obviously, everyone's seen the explosion of all the data centers across the US.
Some states are actually warning their customers the electricity bill is going to go up as a consequence.
But one thing that's coming out of this is that the GPU supply is tightening.
NVIDIA and other major cloud providers are consolidating their power
um there's a potential for small labs risking exclusions um but you guys you know at morpheus
i guess you're you're kind of the solution to that with decentralized compute power
um do you see the the ongoing g GPU supply continuing to tighten?
I mean, at least for the near future.
And that could be a commercial advantage for you guys because you're going to offer decentralized access.
How do you think that's going to go?
Well, the great thing is once somebody buys that GPU and they hook it up in a data center, they still have to do something with it that makes money, right?
And the utilization of data centers is, let's say, 70%, 75% on average.
So 30% and 25% of the time, you know, it's in between running jobs, right?
And one of the key aspects of Morpheus is the compute marketplace, right?
If you've got a GPU, you can hook it into SETI
or Akash or Hyperbolic, one of these GPU networks
and put it on the Morpheus router
and all of a sudden start running an AI job for somebody,
right?
You download a copy of, let's say, Llama 3.3
and that person wants to run a bunch of chat tasks.
Great.
Now your GPU is making money
when it would have been sitting
idle. And so we've already seen this play out, right? If I go pay for an H100 today,
it would cost me, let's say $10 to $15 an hour on Amazon or Microsoft or Google's cloud. I'm
paying $10 to $15 an hour. If I go onto onto Akash today, they've got H100s available from people all over the world,
and I'm paying two to three bucks.
So like the reality is here, if you wanted to spend five times less money on running the same GPU,
then you can do that right now.
And to the credits of the community, they've now made it so easy. You just
go to openbeta.more.org and you can use the API and you push a button and you don't have to
download a node. You don't have to know how the router works. You just say, oh, I want this model
and it works instantly. And they've set that up for free for the community who wants to run
AI for agents. And so, yeah, the tools are getting really good.
We're getting there. And I heard the gentleman that was talking and, you know, the concern about
centralized AI. I have that same concern. That's why I have so much urgency to see freedom of
intelligence that is unrestricted by the state and is owned by the individual. That's what I want to
see. So I'm just going to keep working until that's fully rolled out. Yeah. I mean, we're seeing open source, but closed compute,
I think, you know, is occurring and, and, you know, you guys hopefully have a solution for that.
Sorry, Dirty. I think you had your hand up. Go ahead.
I did not have my hand up, but that being said, yeah.
Well, I think what we're talking about a little bit as well as with compute power, right, is putting it to work.
And yes, there's concepts being used with useful proof of work or proof of useful work in that idle time is putting it to work. And
one way also would be to use it to mine crypto. Another is for compute power to solve problems,
of course. Those are real world use applications being used right now with AI per se. So
efficiency really matters now. And whether it's GPU or CPU based, I think it's a combination of those two, is that it's efficiently operating those systems.
Right now, with quantum computing, right, we just, we don't, we're not there. We've got too much to build. There's not enough compute power to actually build it right now.
So as that's developing, right, you know, what I think integrating of CPU with GPU is the future.
I really believe that. And that also potentially decentralizes that network.
Yeah, I think we're dangerously coming to a situation where computer access is
the new gatekeeper you know if if you know you have these cloud you know where where you know
it's all centralized so i i think you know it's a balance i think morpheus is a great balance to
that and hopefully you know do much more around that One thing I just want to also talk about in the news, we're
seeing, you know, standard sort of things, people interactions getting thrown over into AI. Like,
for example, in the UK, the NHS is piloting a tool to automate patient discharge summaries.
That's one thing. We're also seeing, you know, fast food. I don't know if anyone's been to a
fast food recently, but Wendy's is looking at what they call Presto.
They're joined with Presto so the drive through automation can go, you know, have the AI voices.
You know, what do you think of the trend of, you know, these these consumer touch points suddenly getting turned over to AI?
Do you think now we basically that's going to happen across the board now?
Now, I can speak on my industry. I work in utilities. And I've seen in the last 20 years
such an integration of technology now that it's that whole exponential. We've been taught the
whole time technology will grow exponentially. It is true. It is real. I've been taught the whole time technology will, you know, grow exponentially.
It is true. It is real. I've seen it in my industry.
And so right now, data and access to data quickly and efficiently is really the future, right?
And so, you know, as I'm, you know, now all these data points that I can collect,
I can collect it from my iPhone now literally manage a utility right but you know had to build it over
years but the real I think the future right now is the integration of all the
data and whether it's an LLM ideally it would be AGI in this, but is to do predictive modeling and efficiency-based maintenance, planning,
all that into major utilities, whether it's electricity or water, public works, whatever
you have.
That's the future, is to be able to take all the data points we have now and not necessarily
control it with AI, but use it in your advantage to work more efficiently, because efficiency is the future.
Right. I mean, just specific to that point, IDC came out with a report.
Basically, in 2021, public cloud storage was greater than, you know, traditional enterprise and and now in 2025 they're saying that the data that's stored on the
public cloud is about 100 uh zettabytes which is about 50 of the total traditional data sets are at
30 zettabytes which is about 15 and private device private devices like consumers and iot
is running around 70 zettabytes and i was wondering why that was kind of low, because, I mean, we've all got
photos on our phones and everything. But the thing is, the stuff on our phone gets also stored up
into the cloud. You know, I know I backup my, I have an automatic backup on my phone. So
it's interesting, as you said, once the AI actually gets its fingers across all this data,
both public cloud and the private devices. I think the power
of what we can do is just going to go off the charts. It's amazing what you can, the layers of
data now for, I work in GIS. And for that data, it's amazing the amount of data that we can
attribute to it and just layer infinitely.
But now you have to store all that power on the cloud.
Or, I mean, all that information, that data, right?
And so that's really where the money is, really, is accessing that data and holding it there to access it at any time at your convenience.
So that has to be something we look at as well as how the data is stored and how accessible it is
and how much are you paying for it because, again, accessing data is really where efficiency is right now.
There's so many ways you can save. It's just unbelievable. We can go into a whole nother topic discussion on just GIS and the integration of data there.
Right. So, you know, now with GIS data, you can, you know, if there's a disaster like the hurricane or flooding, it instantly knows where, when and how it's going to have flood across, you know, you know, different zip codes, which I think is amazing.
So you can then bring relief to those or prepare and stage relief to those places, knowing that, you know, it's 10 o'clock or 10, 10.05, it's going to start flooding.
10 o'clock or 10, 10 or five, it's going to start flooding.
Go ahead.
That's all part of the predictive modeling.
Right.
But it's also in a FEMA and event when, you know, I've worked in that type of situation
is that you don't always have time to pull these manuals out.
Okay.
And look through all these, you know, we have all sorts of plans that are required
how to operate in a disaster situation.
Well, that's not what you're doing. You're scrambling, right?
So you need to get in the field. You need the data very quickly so you can make those operational changes, right?
And so to have that literally on an iPhone where I can access all that instead of having to go through manuals and blueprints,
and it's just amazing right and it's
only going to get to the point where as we integrate like I said I don't know if it's going
to be AGI eventually but the LLM still I use that in daily processes to really bounce ideas off of
it so I can make I can decide if I want to use that information that it feeds me. I don't let it make that decision, but in the
future as AGI, if that concept does come to fruition, it's going to be amazing because it'll
only make things more efficient. So that enlides the discussion, is the glass half empty or full?
We look at it and most people right now, they think of it as Terminator, you know, that type of thing where it's going to take over.
And no doubt it will be more efficient, but you have to build the network, or I'm sorry, the artificial general intelligence has to be built in a way that is built for good, right?
And it can be done, but those things are being worked out
right now, if they're smart.
Right, but what we're seeing is where you have situations
that are incredibly fluid, like disaster recovery,
or even military applications where it's very, very fluid,
the ability to have adaptive programming and planning
on the fly is just amazing. Go ahead, Bo.
Yeah, I was going to make two points here. The first being you'd mentioned a company like
Wendy's and basically asking the question, are all of these, is basically every business going to
adopt some form of artificial intelligence? I mean, the short answer is yes, they're going to
have to. It's not going to be a choice. The question, though, is who runs that inference? And more importantly, who has access to that data?
Today, it's not a big problem for Wendy's to say, OK, our CAS register operating system compute is hosted by Amazon Web Services, blah, blah, blah, blah.
is blah, blah, blah, blah. But when they have proprietary agentic AI that are helping optimize
these systems, they need to ensure that only they own that data or else, or at least they know and
trust who does have the ability to access that data because or else what's stopping Anthropic
OpenAI XYZ other super AI company from just becoming the everything company. And the information is going
to be even more important to remain proprietary. You just touched on military application, which
is actually one of my favorite topics in this regard. And I'm just going to keep preaching this
until eventually the Department of Defense reaches out to me and I can help describe to them why they
need to do this. But minimum viable inference for your defense programs is going to be absolutely crucial.
And if I'm hosting my intelligence agency agents and large language models on centralized servers owned by companies, one, that opens the door for a rogue employee to backdoor the information to an adversary. And two, it opens the door for an adversary to physically attack the infrastructure that
allows my defense systems to operate, as opposed to using something like Morpheus, where they're
getting the inference that they need, they're accessing the compute that they need, without
having to even know where the physical infrastructure is being hosted, because it doesn't matter.
And the person hosting that physical infrastructure doesn't know that they're providing compute
to XYZ intelligence agency.
And it creates a much more secure system.
But that's what I think is that building on chain,
your general intelligence, your AI on chain, so you can actually see, right,
what information is being fed into is super important, right? That's part of the decentralization
part, is if you can have it built on chain and have some record of what's actually being built.
I think that's super important. Yeah. So, David, I'd love your input on this,
because I want to make sure I'm not providing misinformation.
But my understanding is that like the data itself, so like the inference and metadata could or could not live on chain, right?
Like the actual agents would have the ability to interact with blockchains, but they don't necessarily have to.
Yeah.
I mean, it's effectively up to you what you
want to keep private versus what you want in public. But what I want public is how the agent
is functioning. Just like Gentleman was just saying, I want to know transparently which tools
it's using and which data source it's using and where it's getting this information, what it's doing.
So if you look at the plans going forward for the ecosystem, I think it's 100% that the
agents are going to end up on chain, just like you publish a smart contract on chain.
I know how Aave works because I don't have to take their word for it. I can just look at the
information and see, oh, the tokens go in here. They get lent over there. I get my interest here.
Got it. Like that's the level of transparency we should have in all finance, right? But we're just
rebuilding finance from the ground up. We have an opportunity to do it right this time in a way
more transparent, predictable, and deterministic way. So yeah, we ought to do it right this time in a way more transparent, predictable,
and deterministic way. So yeah, we ought to do that same thing for agents, just like we've done
that for smart contracts, right? And most of these agents are being built on smart contracts.
So it's a natural extension that somebody develops a new agent, they're going to publish it on chain,
then it's easy to discover, it's easy to audit. And, you know, like we were talking about earlier, I actually think insurance or insuring
of agents is going to be a big thing, because then all of a sudden, the risk goes away, right?
If somebody has audited an agent, and they're willing to insure it, that would be a really
powerful signal to me that it's okay to use that agent.
I was speaking over at the SALT conference earlier this year in Bermuda,
and they're like the world global hub of insurance,
and they're already talking about, hey, we're willing to insure agents,
and they're insuring AI companies.
So that's coming.
Right. Yeah.
One just one make a point in today's newsletter.
There's an article on how AI is going to change a military structure
that really hasn't changed since Napoleon.
And one of the pieces in there talks about special operations forces are already
using AI to lighten the operator's cognitive load, enhance
mission planning, do situational awareness and administrative efficiency. So the military's
already knee deep in AI, probably has been for many years, but at least now we're starting to
see more activity there. Go ahead, Captain Levi. Yeah, so a quick question. I'll go down on the privacy aspects, connecting it to quantum
leaps. And yes, of course, not very many people know it, but there are advancements also made
in quantum computing, you know, saying that people are becoming a certain point to build.
There was a time, a period of time where majority of the media spaces was about quantum proof
technologies, where three projects were bringing quantum proof technologies.
And then a thought just crossed my mind.
There's actually two aspects when it comes to regards of privacy of data.
Oh, okay.
So, in regards to the encryption,
the quantum proof techniques, you know, on all the areas of data from storage to transmission of data, is anything being done? Is there any news so far about quantum proof tech?
Because like you said, especially the military, they've already been using this for a very
especially the military, they've already been using this for a very sufficient period of
time.
And so do you think that individual privacy data could still depend on encryption or will
newer methods, privacy-focused methods be implemented?
Okay.
Rohit, go ahead. Yeah.
Yeah, I just wanted to chime in and talk a little bit about AI agents.
Yeah.
One of the things that is happening is AI agent was such a wonderful name for describing
the agency of AI, but what it has been reduced to comparing with the customer service agents or agents
which can book flights and limiting the whole use case to replacing humans, while it was
much more than that.
And one of the problems of that is that when you're trying to implement agents in a real
world business use case, if you're doing it just on the basis of me too and we should
be on the hype train and adopt this technology,
then most likely you're going to shoot yourself on the foot and which is what is happening.
Just today there's a news of an AI agent in Salesforce was basically hacked by, prompted
injected by security researchers with which they were able to completely make the salesforce agent
spit out the complete database of customers that they had so so prompt injection i don't know how
many people are familiar with that but prompt injection is such a such an open vulnerability
at this space that we are living in the world of what sql injection used to be in web 1.0 world.
Agents are still vulnerable at that level in terms of a prompt injection.
And if you're using a third-party agent and third-party services
to integrate with your proprietary data,
then you're just exposing backdoor to your most priced room in your house.
And it's so vulnerable at this moment that people are already doing jailbreaking to LLMs
just for fun.
But in an enterprise application, the risk is much more higher.
And if you are talking about military-grade applications of AI agency, I would keep my
fingers crossed if people are really understanding what AI agent is while trying to implement and it's not happening for just for the sake of implementing it.
And it won't be surprising. Many people get it wrong. And Salesforce is a classic example using agent for Salesforce.
Yeah, I think, yeah, you're absolutely right. In the last, I've seen in the last three to four weeks, an explosion in the security category of AI headlines coming
out. AI agents are the real key weakness that, you know, people are breaking them. There's been
some black hat activity and conferences also talking about that, that it's really a massive
weak point. Go ahead, Captain Levi.
Cap, I think you're rugged again.
He's off stage.
I do want to just bring up one point.
People have been talking about quantum computing breaking the Bitcoin encryption and all that.
Current estimates are that you'll
need tens of millions of physical qubits, which we're nowhere near close to, apparently. So the
fear of, yes, we need to implement, yes. I mean, David mentioned they've got an EIP coming in
to fix that, hopefully at some some time but it doesn't seem like
we're anywhere close to that at least for the next three to five years but who
knows anyway go ahead right I would say that the same quantum computing can
break encryption is very similar to if people remember early 2000s the Y2K bug
right right said that because of 2000 the dates will roll and all the computers
will crash it's it's almost saying the same thing if you're saying quantum will break encryption
yeah yeah it's going to be certainly a long while i mean yeah one of the yeah well estimates was
no go ahead go ahead bud i was just going to say that a lot of that do dooming over quantum basically comes with the assumption that we just wake up tomorrow and someone cracked quantum and nobody else.
Like, like, like it's just going to happen all at once.
And nobody's going to be to prepare their systems for that, which is, I know, the point you're making with Y2K ultimately.
And that's just not the reality.
Yeah. Yeah. I mean, we might have a black swan event,
like they say, but you never know.
And yes, there was the Y2K.
Apparently there's also a new Y2K kind of problem
coming up in 2038.
So that's something to look forward to, apparently.
You can't wait.
Because in Unix, dates are stored as a signed 32-bit integer,
and it cracks up on the 19th of January in 2038.
So hopefully we'll also have solved that problem as well.
Go ahead, Rohit.
Yeah, no, I would just bring in another aspect of Web3 and agents.
You mentioned, rightly mentioned that when agents we are dealing with
agents it's very important to know what exactly did they do in terms of traceability and somebody
might say from a web 2 world that well you can get logs and trace it but being on chain especially
in the world where we are moving towards agent to agent interaction, those interactions will be much more required,
not just the traceability, but making sure that authenticity is maintained by third party.
And nothing can do better than Web3 in that case.
And I'm sure everybody is familiar with MCP.
MCP is the new API, apparently in the enterprise world right now.
Everybody is doing that.
But at the same time, now we have A2A,
agent to agent protocol proposed by Google.
So the moment that becomes more ubiquitous
and becomes completely adopted by the enterprises,
then having an on-chain track of agent activity
would be a huge use case.
Right.
Yeah, I'd love to bring a real-world example too as to why this is so important, which
by the way, you'd mentioned just a few minutes ago that, you know, making sure these systems
get done right, and the implications of them breaking things.
I'll just tell you right now, stuff's going to break.
Yeah.
Like, people are going to break stuff.
Now, the reason though, to give a real world example of why Oncane is so important
is everyone listening, imagine JPMorgan Chase integrates some type of agent that helps you
manage your money. And now some type of black hat hacker moves and drains every single retail bank
account into a singular account offshore somewhere in North Korea. And they're now being sued by
the FDIC. Let's just say the FDIC. And the only data that we have during that suit is
their own internal logs of what happened. What are the odds of getting to the truth there?
I mean, zero, less than zero. But if it's on chain, we'd at least know what happened. At the very least,
we'd be able to know definitively, this is where the exploit happened. This is what occurred,
and this is how we try to rectify the situation. Yeah, you're absolutely right. And while money
is very important, our whole supply chain is getting shifted over to AI and AI agents. And I think, as you say, if we don't integrate blockchain
or some distributed ledger technology to track this stuff,
some hacker's gonna come in and all of a sudden
everything's gonna stop.
I think we're way past the point where we can go back to,
oh, the computer's not going, oh, I'll just use paper.
I mean, no, everything's just going to
crash and burn if we don't start to solve some of these intrinsic and systemic issues around
security. Go ahead, David. Well, that's why crypto is so powerful, is if I'm holding my private key,
I'm not exposed to those type of systemic failures, right? Who failed in the last bear
market? It was the custodians. It was the people holding money on behalf of others. FTX was a
custodian. You know, Aave is doing just fine. Compound is doing great. Like all the DeFi
protocols that didn't hold any custody, they weathered the bear
market just fine because it was never a risk factor. Right. And so that's like you, like Bo said,
it's going to happen, but it's going to happen to people using centralized systems. And hopefully
they learn from that painful experience and stop using that centralized system and putting their
money at risk. Right. I've come to the conclusion,
if I don't hold the private key, it doesn't exist because property rights guaranteed by
their third party, they always fail. It's just a question of when. And so I've spent a lot of time
building in DeFi and crypto the last 14 years. But I also built some public companies and they were all at
the mercy of regulators and exchanges and a bunch of third parties. Right. And guess where I'm doing
better? I'm doing better in decentralized systems. Even if you can get all the way to the, you know,
the end of the golden rainbow and, you know, have a public company. If you've rooted it in the state,
you're still vulnerable to all those
centralized systems. So yeah, people are going to learn and they're going to learn fast. But if you
want to protect yourself from that, if I hold the key, then I'm not vulnerable to that. And that's
why I need a personal AI that can only use my key in the context of what I want to accomplish.
And to their credit, Coinbase has done some really great work on this.
They have a developer kit for agents, and they've released entire tools just focused
on, I want to give my agent a $100 budget, and it can only use this app, and I'm authorizing
it for the next month, right?
Those kind of granular controls that you want over your money, that is
now live. Their MPC, multi-party compute implementation is great. And that's what
most people I know are using. Yeah, I think while it's really important that we have wallets with
the security for currency and value, we need an equivalent wallet for the data, right?
I think we need cryptographic access with the private keys for access to data, not just
the money now.
And I think that's becoming really critical when we see AI agents getting integrated into
access controls and a whole bunch of other things where you have massive potential security
vulnerabilities that it's good for me to lock up my money but I
also want to be able to lock up my data go ahead dirty so you know decentralized
assets right your investments is a double-edged sword what is more likely
scenario give you two scenarios. One
is that you are on a centralized exchange and bad tanks and you lose everything potentially.
Or that you are on a non-custodial wallet and that your private keys get hacked or you lose them,
right? What is a more likely scenario? So not that maybe it's not one size fits all.
Is that a combination of those two things?
Maybe a better answer is, and that also gives you maybe a little more protection
and to be a little more diversified, I think,
because right now it's a bit of a wild west,
and there's not enough education to bring traditional finance right now into the market because they don't have the education
to actually do anything safely right now until they get that education i don't think it's ready
to just everybody to come on over that's just my opinion yeah go ahead bud um so i actually i very much agree with this point
actually but there are some solutions in place so for example multi-signature wallets are a huge
step forward in this direction sure and basically for anyone who doesn't know what that is it would
be the idea of having a wallet address that let's's say it's a two of three, meaning there are three other wallet addresses tied to it. And in order to move assets in and
out of the wallet, two of three of those addresses would have to sign. The idea being you own one of
the wallets, you have trusted third parties who probably don't know who each other are,
owning the other two, which means that only you would have the ability to move the assets, even if you lost your private
case, or I'm sorry, like access to the wallet. And so this is
an issue, though, that I've heard the callic theater and speak
about a bunch of times is trying to figure out what is the best
solution to this in the future, and like different social
consensus options for your own your own keys basically and the dai is going to
go through the same exit and i made it a two or three i might trust david for example with
access to one of those keys david on his own could not execute my agent but he could help
me recover it should that scenario happen the difference being
i think we lost him
no did we lose bo yeah i think we did but can you repeat that last part i heard
yeah you you just like last thought if you could repeat oh am i am i back
am i back yeah okay i'm sorry i was just saying um the the major like this is definitely something
that dai is going to have to tackle as well but i you know i might trust someone like David with a key to my agent, and he wouldn't be
able to execute on it on his own, because it's a multi signature, but he would have the ability to
help me recover it if I needed to. But the important thing there is that, as opposed to
trusting a centralized entity that has no personal connection to me, I am choosing my own trusted
third party that I think is best for me, which is a really
key point of differentiation between blindly trusting a large entity that may or may not
have your personal interest in mind.
I mean, they don't have your personal interest in mind, put it that way.
Yeah.
Dirty, go ahead.
I didn't have my hand up, but...'s still up great point there great point there though
and let me just touch on that with the multi-six again the most traditional finance
investors aren't really educated enough they just jump in here too early and they're looking to, you know,
be invested into DeFi, but there needs to be education. It's super important, right? And so
we need to work on that first. Yep. Totally agree. So we're coming up to the top of the hour.
I just want to share with people i'm going to
post to uh the mobi media account um a link to the newsletter that i do five days a week
uh for anyone here uh just click on the subscribe link throw in your email and then leave it and
i'll give you a one month subscription to the newsletter so you can see on a weekday basis what's happening in over 31 different areas, categories like military,
healthcare, finance, and a whole bunch of other stuff.
And then on Fridays, I publish what are the insights for the week.
And then at the end of the month, I publish what are the insights from all these headlines from over the month.
And I also do that on a daily basis.
So that helps all of us to keep up to date on what's happening in this amazingly fast
moving field of AI and treat it like a smorgasbord.
Just cherry pick, go to the one you want.
There's maybe 15, maybe 20 headlines in there.
That'll get you totally up to date on what's happening today and over time in artificial intelligence.
So thanks Noah. Thanks for hosting us today. Really appreciate it. Thanks to everyone listening
in and we'll see you again next week, next Tuesday, same time, same place. Thanks so
much. Thanks David and thanks Dirty. Thanks Bo. Thanks Levi. Thanks everyone for chatting
in there. We'll talk soon.
Take care everybody.
Thank you everyone.
