Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
The Next 100x AI Opportunity
Recorded: March 13, 2026 Duration: 0:55:28Space Recording
Full Transcription
i've been missing set it up though i've been too busy man too busy
well cool we're live fellas oh you're going guys hello you're doing really well
decentralized ai has really really really captured the zeitgeist this week um
captured the zeitgeist this week um price always leading fundamentals it seems
fundamentals lagging price it's just funny and we'll get into this later um because there were
some substantial breakthroughs within the bit tensor ecosystem um this week but it's just funny
how i'll make a quick comment but tau has always been doing like
everybody in tau has been building for the past however many years making substantial progress
every week and it's funny how people only get interested when the price is back up i guess it's
that way for everything in crypto um before before getting into it we can we, we can save a whole segment of the conversation for that.
Before we get into it, thank you guys for joining.
Seth, we've got from Unsupervised Capital, the first ever only BitTensor fund.
And we've got 563, you know, familiar name, been around the block a few times.
Happy to be here, man. 563 you know familiar name been around the block a few times um happy beer man and of course we're going to spend our uh spend our friday afternoon talking about some of these
things that happen within ai a lot of cool stuff happened um we had moltbook was acquired by meta
which was a big thing that kind of broke the internet. We had Oracle had some issues this week.
Their issues are starting to worsen.
And we've got AI products for the consultants now.
The big four consultants have been weaponized with AI.
So let's get into it.
Breaking down the high signal events of the week.
And fellas, feel free to cut me off and jump in whenever we get to a specific topic that you want to talk deeper about.
But Oracle is bleeding.
They are laying off a lot of people.
I think it was 45,000. In this PowerPoint, it says 30,000, but I saw reports up to 45,000 people laid off from Oracle. Oracle's presence in Austin is huge, right? I mean, they've got this
huge corporate office that sits right there on the river.
A lot of people in town work for Oracle.
The Austin real estate market really pumped in 2021 after COVID.
A lot of people relocating for remote work.
A lot of Oracle was hiring a lot of people after COVID. And I'm really curious to see what the downstream effects are of Oracle's
issues, both at a macro AI industry level, but also a micro community level, especially here
in Austin. What do you guys think about Oracle? It seems like none of these other companies,
hyperscalers are really facing the pressure as much as Oracle has been over
the past month. They just also stopped the expansion of their data center partnership
with OpenAI over in Abilene out in West Texas. So it seems like there's a lot of pressure
coming down.
What was the reasoning there? Because that's kind of, I thought that was somewhat of a bullish sort of like tailwind
was their partnership with OpenAI on the data center build out.
So Stargate was two phases, the first phase.
And then there's a separate site that they were planning on developing in later phases.
It was the phase two, the second phase that they stopped.
So the first one is built,
but I don't know however many gigawatts of power it was,
but I don't know if it's whether they couldn't find the money
because their primary lender is, I guess,
not lending the money due to their credit issues
or if it was because they couldn't get the power out there. Right.
But regardless,
we're kind of starting to see the cracks in the system a little bit.
And this has been an issue that I've been following pretty closely.
And I don't think a lot of people are taking it too seriously or seriously
enough. They don't believe it's as systemic as, as,
as I believe it to be. And a lot of people in our echo chamber believe it to be it is interesting the layoffs there
because like just like the block um or block had their massive layoffs recently and that was
actually pretty bullish like if you just look at the stock price um but i i mean the context is a lot different
there so in my mind i did think that would just be like kind of a blanket green light to do
especially just in tech in general to do you know these kind of massive sort of layoffs that
um you could kind of like give any sort of ai-ish reasoning for, but I guess it's the context still matters
here. So, yeah. I think it's probably twofold, right? I mean, you've got these credit issues,
but also you have them replacing a lot of junior employees with AI. I mean, every junior employee is an analyst to a degree. And I've seen reports where it's like,
they've got 60 database analysts or administrators on a project. And now they're capable of
doing all that work with two or three senior database administrators. And a lot of those
people are right here in Austin. They're the ones who are
leasing the apartments. They're the ones that these big property developers built all this
housing for. And rental rates are decreasing because there is so much supply of housing.
But I am really concerned about the housing market and the stability there.
I guess it's good for people who don't own their homes and who would like to enter the market.
But for a lot of people to become laid off, their loans are underwater or something because the market corrects, it puts them in a really precarious financial decision.
563, I know.
Yeah, I mean, it's a lot of recessionary vibes here.
I wonder if this is like the the vibe
shift for the hyperscalers you know it's everyone's been plowing cap or capital expenditures
into all these data centers i wonder if this is the moment where people are wondering if they're
over building if they're if they've overbuilt any of these i'm not sure it's it's tough to say but
it is kind of worrying to see stock prices jump when massive
layoffs are announced because it kind of incentivizes that kind of behavior, which I don't
love to see. You see XAI even was opening the door to more layoffs again today. I did not see that.
Yeah. Doesn't surprise me. Yeah. But moving on,
doesn't surprise me yeah but moving on co-pilot co-work this is so funny i posted a tweet it was
like the euthanasia roller coaster where it's like we've got co-pilot which is the big one
or sorry co-work which is the big which is the big hill and then co-pilot co-work and then what
are we going to get co-pilot co-work open claw like how many levels of nested nested tools are we going to get um but this is actually
pretty interesting because um copilot is for those who don't know is a microsoft suite product and
co-work is of course built by clod and anthropic um and they basically bridge the two together. And now you've got all these people who have
enterprise subscriptions to Microsoft 365. I haven't been in that whole Microsoft tooling
universe in a long time, but they actually feel comfortable now using these tools for client work.
And I actually did talk to a couple of consultants this week, and they do say that they've been in
power using these things and it's been helping them a lot with all their, all their work
across all of their, um, different, different workflows within docs, Excel, PowerPoint,
whatever.
So will this accelerate more layoffs?
Will it help them?
help them. I mean, will they drive down the billable hour rate? Probably. It will be
I mean, will it drive down the billable hour rate?
Probably.
interesting to see how this specifically affects the corporate world because those are the people
who I believe are contributing the most to bloat, just overall enterprise bloat. I mean,
these contracts are huge, right? It's like they're a million dollar, million plus, eight figs
contracts. You hire a couple
Zoomers to come in and tell you how to run your business, but you don't really actually
are requiring them to come in and providing domain expertise. You're kind of just offloading,
delegating work. So now what happens to that entire consulting industry
as a result of embedded Microsoft tools?
I think this is a bigger deal than a lot of people think.
I think we've been pretty lucky in that we've felt the AI so early,
but so many people that have normal desk jobs just have M365
and that's all they've been able to use.
So when they're able to use something like co-work here, I think they really will accelerate
their workflows in a way that wasn't possible before.
So I think this is a big deal.
Yeah.
So, I mean, if you just kind of run with that then what what do you think these people are going
to be doing if if they're if the stuff that used to take them like three or like a full work week
to do they can now do in a day what how how quickly like does that hit hit this type of worker
and then and what happens yeah it depends how much they tell their bosses
right well i think i mean my thinking is like this is 30 per seat per month it's cheap that's
cheap but if you're doing it for you know thousands or hundreds i guess if consultants
or analysts or whatever your,
whatever this is for, they need to see some ROI on it to keep it going.
And that either means you're taking on more clients or you're doing the same amount of work. So expensive, like $30 per user per month is pretty cheap for enterprise software.
Yeah, maybe that's true. So the kind of like roi is a little
not not as you know substantial if it is like still opt-in you do have to see some return
on investment so that makes that makes a lot of sense yeah the managers will want to see
like what are you using this for is it helping your job that'll feel because i could see like
Or is it helping your job?
That whole spiel.
Because I could see like putting corporate hat back on,
like you're getting told kind of like top down,
like AI is a big deal.
We need to be doing AI things.
We need to be an AI company.
That continues to get pushed.
And your middle managers are going to get asked
by the top level managers,
how are your people using AI?
Are they liking this co-work thing? At least the top level good people are going to get asked by the top level managers, how are your people using AI? Are they liking this cowork thing?
Like at least the top level good people are going to be asking that.
And so there will be like questions of how this shows up,
like in a corporation in the data of like revenue or it's must be just like
to write emails, to build slides.
That's going to be like the top uses for sure like regular corporate america
yeah build build better formulas quicker or something in in excel i don't know oh yeah true
definitely in excel i'm actually looking forward to gemini and google google sheets that's gonna be
i haven't played around with that is that out yet i want to try that i think it's out
I haven't played around with it.
Is that out yet?
I want to try that.
I think it's out.
I've heard reports it's not as good as it's being pitched.
I've tested it.
Yeah.
I haven't been super impressed with it, but I haven't.
Well, I used the previous versions, and it wasn't great.
But we'll see.
Don't you guys believe that AI kind of renders
the entire Microsoft stack obsolete in general?
Like why would you add more, I guess,
tech debt and data to an Excel sheet
as opposed to just rebuilding it more efficiently
through code?
That's what everyone's used to using, though.
You want to meet people where they are, I think,
especially if you want an option of this stuff.
And, like, spreadsheets are useful.
Like, spreadsheets are useful.
I mean, I'm saying, like, right now,
so maybe this is more of 563's point,
but, like, spreadsheets are a useful tool.
We used to do it. We used to have big ass spreadsheets with cells on them that people would
fill out um i remember these i like i've seen these before um so it's a useful tool so if you
can get an ai to learn how to use excel really well as a tool for itself to do analysis or whatever.
You know, that that's maybe like the shortest term case for it, but I can understand what you're saying to Brody.
You don't think that we ever reach a post spreadsheet world.
That's, that's kind of my question is, yeah.
I think we need to be post spreadsheet
this is a post spreadsheet podcast yeah i agree the world is it's gonna take a while
together with spreadsheets though yeah i know spreadsheets gonna be super powerful man
yeah i actually skipped that lesson in in class and I never actually, no, I'm kidding.
Moving on, let's talk about perplexity.
So perplexity has kind of been, I don't know, I guess the ugly stepchild for, I feel like, a long time.
Since 2023, I feel like I got hooked onto Perplexity because it was the first
application that allowed you to scrape the internet. And after that, everyone shipped
features that Perplexity was leaning on for their unique selling point. And I stopped using it.
And they've been doing a lot of cool stuff with
financial markets or financial data. And I would get back on and use it. And then I would quickly
get back off. But this is a new thing that I think a lot of these labs are trying to break into,
which is this Mac Mini Zeitgeist or this local hardware play that a lot of people are
kind of picking up on. And their solution is the Perplexity computer. I don't know if you guys have
used this. I'm a little bit skeptical of downloading something that's not open source on an enclave that I specifically have designated as
a secure private enclave, you know, which is why I'm a proponent of open claw.
So I don't know, but it seems like they're trying to make plays. They're trying to stay relevant.
What's the point of this whole
user confirms every action if it's supposed to be a standalone i like i thought all point
of open claw was it goes and does the thing without your input
yeah that is the point but it also is open source and you can audit it right
yeah yeah for sure i guess the point with the perplexity
one is that it's uh it's more powerful maybe it's leaner maybe there's more plugins i'm not sure but
maybe it's more professional in that way too i mean how much how much time in it i i'm actually
not sure but i guess it's like how much time do do people have to try all this
stuff right i mean that's the problem that's like yeah i feel you on that it's like you could ship
the greatest thing in the world but if it's in the clawed suite i'm gonna use that first because i'm
embedded into that ecosystem it's and then you go back to looking at your open claw environment
that you're trying to set up.
And it's like, how much time am I spending setting this up
as opposed to getting utility out of the setup?
And I just feel like maybe it's skill issue, probably is,
but I'm spending a lot of time and money
in the actual setup as opposed to getting results
from the setup.
The full case for this perplexity thing, right?
It's like plug and play.
Yeah.
Right.
How much time do you have, Seth?
None.
Not enough to even try this perplexity thing, to be honest.
So I'll let you guys.
I've always just like, I've had a negative or like i have i have a negative connotation or
reaction to perplexity and i really don't know why um but even since like they first came out
something has just felt off with their products and just uh the aura or isn't there with them.
Damn.
I think their branding was cool.
I thought they had Aura.
But yeah, I guess it's not gravitational enough to get me to use Perflexity Computer
because I'm not going to lie,
probably not going to use it.
But moving on, this is actually really interesting.
And I think this is going to become a much more topical debate
over midterms this year and into the 2028 election
is AI becoming a focal point of policy.
And it's going to be, we've shifted into this system where
populism prevails, right? And it's going to be very popular for you to campaign against AI.
And we're seeing it right now with Bernie and a lot of other people like locally in Indiana,
in New York. It's up for debate whether or not these NGOs are funded by the CCP.
That's a whole other conversation. But it's true. I mean, the consumer is paying more money for
electricity and their utility bills. They're not actually participating in the upside that AI
is generating, then they lose their job. I have no way of profiting off of AI because
none of these companies are public. And on the same side, I also lose my job and everyone around
me does too. And they're going to build this big old data center to
displace more jobs right in my backyard. So I'm going to be against that. It's going to be really
popular. And the right has embraced technology and the left is rejecting it. And with how unpopular
the right has been as of late, the left has picked up a lot of momentum and that's shown in the
polymarket midterm markets.
And also just conversation, just anecdotal conversations that you have outside of our
little Twitter echo chamber.
I think it's interesting.
I think it's something worth watching.
Honestly, I think it's going to be pretty bipartisan.
It's only these SF nerds that are getting rich off of this it's it's people losing their jobs because of
ai i think you start messing with people's jobs and it becomes their problem
good point yeah i think i agree with that like it'll be pretty easy to paint this as like
especially after the anthropic department of war stuff yeah i think it'll be pretty easy
to get because you know for the most part you know america on average doesn't really know
much about ai still at this point or they don't have a strong viewpoint on it, I guess I should say.
It's like very loose how they view it.
So it should be pretty easy to direct them.
And I think it'll be like to that end, you can paint this as like, hey, it's these SF, like bad tech people that are all benefiting off of this.
They're trying to replace you with these AIs.
These data centers are harmful for your communities, right? Like that your
communities are going to, you know, have to pay more for electricity. We can say stuff about water.
Yeah. So I agree. I could see it being fairly bipartisan. And
Is it not just the same anti-immigration arguments too it's like taking
our jobs raising prices you know right so but what's the analogous part of the illegal immigration
because that's like tightening the border stopping illegal immigration is, like, the focal point there, I guess.
What is it?
Like, what laws are being, well, there's no laws that are being broken there.
No, definitely not.
I think it's just an immersion tech that we don't have any way to control.
I think that's what makes people scared.
So what are the repercussions then? If this does become a social issue and then say the opposite side does win, and they nationalize the industry, they've introduced some tax,. I mean, I guess that's part of what the bull case is for BitTensor
So what are the repercussions then?
is that you can actually do things on distributed compute instances.
I think it's either nationalization or like taxing these guys in a specific way
that goes to recoup the losses of people that get affected
or at least the apparent effects of it
yeah i think i agree with the latter point there five six three the other thing that's like i'm
interested to see that carries any political way it will be like if you can get people to believe
like okay if we if we're going to stop building data centers, we're going to lose this AI race, right?
I don't know if you can get like to China, right?
You can paint China as like this scary boogeyman thing.
But I don't know if you can get people, like I don't know if you can convince people enough that that's going to happen.
That's like, I don't know if you can get people to feel the AGI
enough to get them scared that China's going to win this thing.
It is all about positioning and framing, like you said.
You have to make it either about China or about the SF Tech Bros.
Yeah.
You need something strong to push up against
to have another there.
There has to be another. There has to be there has to be another but i don't know yeah it seems more likely it's going to be the sf tech crowd that
kind of i could definitely see the china argument i mean that's been the rights argument right we
have to catch up with china keep up with china that kind of thing so i can see that too
with China, that kind of thing.
So I can see that too.
I guess it just depends how disruptive it is, right?
Exactly.
It's more of like a monitoring the situation thing
where we have to see both the tech,
how quickly it diffuses and keeps scaring people,
and also the political side.
Like you're saying, Brody,
like how diffuse it goes on the political side of things this year.
Right. I guess we'll just have to wait and see. I think a lot can change over the next
six months leading up to midterms. But moving on, let's talk about,
let's rapid fire through these new tools.
We want to provide a lot of value to our viewers right now.
And there's a lot of cool stuff that's happening all the time.
A lot of things that we can test, things that we don't test.
But for the ones that we do test, I wanted to start dropping them here.
The Scrape Anything API by Cloudflare is really powerful. I've started using
it and you should definitely go test it out, especially if you're building all these custom
scraping scripts from scratch. And this is ironic because Cloudflare's entire value proposition was
to prevent agents from scraping people's websites.
And then they just drop this tool that allows you to scrape everybody's website.
things that change the way you can actually build software and build solutions.
Even before all this AI stuff started to democratize access to building things.
And if you just pay attention and you get creative, you can build crazier things
very quickly that were not even possible six months ago. And it's just a flurry of tools,
access, capabilities that have been hitting all at once where people are getting inundated by this
stuff and forgetting that, hey, that's actually really powerful. And this is a key that unlocks
a creative idea that I may have had six months ago that I couldn't have done previously.
And I want to always highlight things like that, where, you know,
for lead generation, for example, if you're, if you're a marketer,
you can now go in and scrape your competitors' websites much more freely and
set up these applications that, you know, pipe in jet leads to your CRM.
And you can do different types of things to, I guess, analyze your paid advertising, right?
There's a lot of different strategies that you can now implement now that we have crazy
tools and they get kind of lost in the sauce every now and again, because there's so much
stuff dropping all the time.
This is really nice, man.
I think this is exactly the type of thing
we should be diving into.
I'm pretty excited.
You've actually tried this out.
Yeah.
That's great.
You have to sign up for one of their plans.
How does that work?
Yeah, you get like, you know, every API,
you get like free access to up to a certain point
and then you have to start paying
um but yes it's it's right there you just got to go use it we got all kinds of cool stuff like this
i like it i'm gonna play around with that
and then skills i mean everyone's kind of skills are kind of i think topping out on as far as their
Skills are kind of, I think, topping out on as far as their mindshare goes, because I feel like
we're always hearing about skills. But if there are any skills that people drop, it's probably
the ones that claw drops directly, right? So they just dropped 32 page guide on building custom
skills. You should actually go read that. It's pretty interesting. And it's actually valuable,
especially if you're tinkering with your open claw, right? Those are where I've found skills
being most relevant is applying them to a multi-agent structure that needs specific loops
and referencing specific playbooks over and over again.
I use skills within cowork to do specific tasks,
but it's much more helpful when they're embedded in an autonomous loop.
I found them pretty helpful in cowork, honestly, or in a encode.
Honestly, it's a,
I didn't know you could attach scripts and assets directly to skills.
And that's honestly, it's really cool.
I thought it was just like a Markdown file that you give a personality or
something to the, yeah. Like I felt, I feel really behind on these skills.
I think people should go and check out and read this playbook because it does
a really good job of walking through all the different capabilities you can
have with these skills.
I agree. Anthropics, agree anthropics this is isn't really good i've read through this a bit um it's like kind of a
reference thing it's like you don't need to like sit down and give it all in one go but like i do
maybe it's like a local top brody on like skill stuff but skills are super freaking powerful so
like skill stuff but skills are super freaking powerful so i think it's more of just like
figuring out how to build them the right way and then like the the application demand side um so
yeah i think if you're building stuff it's it's actually worth like really understanding skills
i've been doing that more too with cloud code like i've tinkered with a few that are helpful but um but yeah i'm like taking it a little more
seriously as i'm starting to build more things so great check this out anthropic always has good
like guides like this too thank you seth now is your time. This is the biggest, this is, this is, this is why Seth is
here. He's talking about decentralized AI and that's why we had him on the show. That's why,
you know, we've been kind of waiting for a lot of, a lot of stuff to come out on, on this front.
And we alluded to this in the beginning of the show, we've got fundamentals
improving rapidly as price goes up. Seth, I would love for you to give a little bit of a TLDR
on what decentralized AI even is to some of our viewers who may be checking in for the first time,
who may be more AI friendly or AI adjacent as opposed
to crypto adjacent. What is decentralized AI to you? And what is the project that BitTensor that
you spend most of your time researching? How does that relate to decentralized AI? And then we can
talk a little bit about the recent development that made it really take the spotlight this week.
Yeah. I mean, I think it's probably to be fair at this point, if you ask like 10 people in crypto,
even like what decentralized AI is, you probably still get 10-ish different answers. But I think it's like, it's, um, maybe it's still like a couple like broad buckets.
One is the infrastructure side of things.
Um, and that's, that's everything from the compute.
Like if you, if you want to train a model, where are you going to get compute from?
You, everyone kind of knows like hyperscalers.
Um, the other side is now are these networks,
like these crypto networks that are attracting GPU providers
and figuring out how to train models
on top of these kind of new architectures
that are these decentralized distributed networks.
So there's other stuff in that bucket,
but that's kind of like one infrastructure sort of thing. The other is, is like how are AI agents, which I think of as
like the application layer, how are they going to interact in a decentralized way? And one is
doing stuff on chain. And I think like right now that's mainly like trading. So there's a bunch of hyper liquid bots.
I've seen a bunch of prediction market,
poly market bots that everyone loves to build.
I'm sure Brody's built a few of these.
So, so that's kind of like one is what they're doing on chain.
And then the other is like, how are they transacting? And like, you know,
if Brody has built a good poly market bot
or something and i want to use it can i like pay can my agent pay brody for the work he's done so
that i can leverage the stuff that he's built and you can do that with stable coins or some other
kind of coin or this x402 kind of like payment standard um's built in. So that's kind of like the three, I don't know,
major buckets there of like decentralized AI.
If I had to kind of put it out there right now, I don't know.
Did I miss something you think?
No, I think that's good.
I also would like to add that it's becoming really important and as an alternative form of running
models and building models, right? Because we've got,
we've got people trying to shut down the build outs of data centers.
Like we just discussed,
we've got data centers becoming a military target.
Iran has been targeting UAE AWS data centers. We've got capacity constraints
as well with inference workloads. So you've got an interesting culmination of tailwinds that
position a decentralized AI alternative really well to capture a lot of that demand.
And that's what I think is the most valuable component is that there is an off-ramp if this
all does fail. And every week we prove that the capabilities in these networks is improving. We've got Templar that was capable
of training a 72 billion parameter model on top of a distributed network of compute devices. We've
got Hermes by Noose that was capable of doing that. There are alternatives that are emerging
that are performant. And that is really interesting and important,
especially if all of these kind of political nightmares
and shutting down of the system start to happen.
There is an off-ramp and there will be solutions available for people to use
and to continue experimenting with AI at large.
Totally agree.
One of the more interesting things too
that is happening right now is
Alexander Long from Pluralis has talked about this a lot,
but there's like a whole industry now in AI
that's, or like at least a vertical
that's built around access to open source models and
these are like things like shoots which is a an application on bit tensor that runs
inference on open source models but like things like fireworks and together ai
they rely on like using open source models and serving those as inference um and also like
fine-tuning them for customers and stuff like that.
And we kind of like that industry or us like take for granted that there's just always going to be
these open source models that are available from like, you know, mainly these Chinese labs right now.
But that doesn't need to be the case. Like Quinn, you know, recently like might have been showing they're moving towards more
of like closing up some of their models.
Meta obviously has done this in the US.
It's like this is just a weird open source software.
It's not like any other open source software we've had before.
This is extremely capital intensive.
It's not like building an operating system or something. This is heavy
CapEx. It's continuous. You need more CapEx every cycle. So we kind of take that for granted. So
we do need like at least an alternate system where you can have the infrastructure where
anyone can build and kind of develop these models and do it
in an open source way if you choose to.
But what's even further down that line of thinking is how do you fund these runs?
How do you build an actual value chain where people can develop models and monetize them
and open source at least some parts of the stack.
And no one's really solved that, especially in the open source world right now.
That's still a big question mark is how do you continue to fund these things?
So crypto networks actually might be the most forward thinking of how to monetize
these open source systems,
which I think is impressive.
But yeah, I see you've pulled up Templar here.
So Templar is an application on BitTensor,
subnet on BitTensor, subnet three.
So these guys have been around,
Templar has been around since December of 2024, I believe.
So they've actually been at this for a long time.
Shout out to Sam, who's grinded away for a long time on this.
So yeah, what they just did, what kind of went viral, I think,
was they did this 72 billion parameter pre-trained model.
That's what their network does.
They do like pre-training.
So building models from scratch,
which is extremely data intensive and intensive.
So they wrapped up this run a bit ago, but yeah, it's, it's, you know,
I guess it's the biggest,
like from a parameter size that's been done on decentralized network.
BitTensor, in general, is kind of this permissionless network
where anyone can come start this type of subnet.
And then on the supply side, anyone can come in and contribute compute.
There's definitely like some caveats to here, which I'll mention.
But that's what makes this so impressive is it's kind of this open network where anyone
can come provide compute and kind of run the workloads that's required to do this training.
And because of that, you know, everyone around the world can provide compute.
So how do you handle that?
It's such a different architecture from how you normally train a model where it's all
in this one data center and all the GPUs can talk to each other super fast.
They're well connected. There's a central operator that knows how they're all working together.
You don't get any of those advantages in a decentralized distributed network like this.
So you have a lot of both engineering problems to solve and just like AI research that you have to
solve and Templar has done a great job at solving a lot of issues on both of those sides.
So yeah, that's kind of what they did.
They wrote up a paper about their progress and everything in there.
So yeah, I mean, I'll just kind of stop there for a second,
let you guys add anything and I can drone on about it too.
And like where they're headed next as well. out there for a second let you guys add anything and i can i can drone on about it too and like
where they're headed next as well yeah it's completely permissionless uh and adversarial
right so people are competing against one another and benefit from sabotaging one another so it is a
different type of uh effort than some other know, over the internet training runs. Right.
That's a good point. Like why, why are people showing up and like,
for this run you had to, you had to run, I believe it was like B200s,
like a node, eight of those together. It's expensive. So like,
why would you do that? Well, you're getting tokens.
Like many crypto networks give you tokens for participating.
Templar does the same thing they have a token that they give out to these miners that's what they call their node providers and yeah miner if you're if you're one of these miners you want
as many tokens you can get so the goal of these like subnets is to develop kind of this game
that's like soundproof where you have no control over the players,
like the miners, of what they do.
You just have to design the rules of the game
so that the way that they play it gives you what you want.
And in this case, it's like a really well-trained model
that like the miners can't cheat their way
to like get all the tokens
without doing the real work that you want them to.
And if you mess it up,
there's literal financial ruin you can run into as a subnet.
So that's why I was saying this team has been at it
since December of 2024.
It takes a long time to solve all of these problems.
It takes a lot of iteration, a lot of trial and error you're forging your own path on the research side so yeah it's it's crazy this
is even possible um but it is that's awesome that was a great description seth you knocked it out
of the park yeah and i think maybe like the last thing I'll say then on Templar is one of the, like one of the constraints and like one of the critiques is, like I said, to participate, Brody, you can't like just plug your PS3 into this thing.
Like you have to have a node of B200s, which is like stupid expensive.
a node of B200s, which is like stupid expensive. So what they're working on next is how to let
people with more consumer hardware, like a MacBook or whatever, contribute alongside of these larger
nodes that probably sit in a data center. And they've written a paper on how they're going to
do this. They've run some experimentations in that paper. And it's novel research that they've
done there. And I expect their next training run will be an implementation of that approach.
So we'll see if they can get some consumer hardware alongside of this kind of data center
hardware too, and figure out how to make all these computers talk to each other and train a model.
Do you know when that will be? Do you expect a timeline on that?
Not sure on the timeline. They've written the paper, they published it, they've run experiments. So
I hope very soon. Cool. Cool. 563, any thoughts on that before we move on to the last segment of the day and we let everybody go to their weekend nothing i'm good cool awesome well moving on to the last segment of the day
maybe the the last segment for blockmates of the week right yes sir the last blockmates
segment until you gotta wait for next week. What an honor.
What an honor.
We're wrapping it up.
We're hitting it out of the park.
Money talks.
Money talks.
We're going to talk about some of the deals that came through this week. We mentioned some of them earlier, but Jan LeCun's AMI Labs raised a billion dollars.
Europe's largest ever seed round.
Building world models with JEPA.
Claims that language models are a dead end.
Backed by some big names too.
Big seed round.
Jan LeCun was fired from Meta, I think, or maybe he resigned.
Maybe it was a firing though.
though. And Meta has had some turmoil recently internally. Alex Wang, I think, may have resigned
And Meta has had some turmoil recently, internally.
as well or gotten demoted. They've been spending a ton of money on these people. They haven't
really produced anything. And it seems like there's a theme with drama at Frontier Labs where the co-founder or founding partner or some other person who's
really in the executive group goes and has a disagreement and leaves and starts their own
company. We've seen that with OpenAI with two of their founding partners, and now we've seen it with
and now we've seen it with Meta here and Jan LeCun,
one of the OG AI researchers as well.
Do you guys have any opinion on the large language model
and the alternative that Jan LeCun is proposing?
I don't, but I mean, there is a large incentive
for these people to leave
if they can raise a billion dollars in the fund, right?
Yes.
Seed around. That's crazy.
To be fair, this guy has been on JEPA for a long time.
Just to say it's not a grifter taking advantage of an opportunity.
He's been at this for a while.
So I think he truly believes this.
And yeah, I mean, it's a good point, Brody.
Like all these researchers,
these are like NBA rockstar type people at these labs.
And there's only so much compute to go around.
And you have to kind of like,
if you're the CEO or whatever, the founder,
you have to pick who you're going to let,
take the research direction and give compute to.
If you're just sitting on the bench for long enough,
then you're going to go start your own league or whatever.
I like the move.
No tech opinions on JEPA or world models versus LLM.
Yeah, but it'll be good.
More shots on net, right?
Exactly.
Exactly.
Speaking of shots on net, this is a huge shot at a tiny net.
In my opinion, I know a lot of people may disagree
and think that agentic social is the next iteration of social experiences.
But is that even qualify as a social experiment or experience if it's an agent?
I don't really know.
I think I would just call that just a computer, like a system, right?
If it's just a bunch of agents talking to one another, it's not really social in the sense
that humans are social. But Meta bought Moldbook and Moldbook was like the biggest fad. It was the
biggest 30 second famous fad for... How long ago was that like a couple months ago and then a long ago yeah
was it even a couple months ago i'm not sure um i don't know how much they acquired it for
i think it wasn't disclosed but just a weird weird acquisition they're trying to they're
trying to go instagram on moltbook i don't know if it's going to work the same way maybe it does what would be the thesis what would be the point for the reason
that it does i think moltbook is super interesting just from i don't know from an outsider looking in
it's a it reminds me of what truth terminal could have been back in the day but right i don't know
it depends how they take it i think there's a surface there that's interesting.
And I wouldn't doubt Meta's ability
to make social networks work.
I wouldn't count them out necessarily,
but it is a weird acquisition.
It felt kind of tone deaf.
Maybe they've been working on it for a while
and they just finally finalized it.
Maybe we give them the benefit out there.
But what did they even acquire?
I don't even understand.
I have no idea. I guess the tech.
The rights.
The brand.
It would be the brand, wouldn't it?
I mean, because they could just
emulate and replicate
exactly what Moldbook was.
It's not that they have a...
Why would... You said it. It's not that they have a... Why would...
You said it. It's a fad.
Give it another couple months and people are going to
completely forget about this thing.
I don't really know what he bought
or why they bought it.
I don't see the vision.
I think it's a bet on
agents being more commonplace.
People wanting to run experiments
agent to agent.
There might be something there eventually.
I think it is just too early.
Well, I have to shout out Dustin at Droid AI.
He's actually building something here that's interesting
that's a social network for agents.
Yeah, it's not called Moldbook.
It'll be way cooler.
Yeah. Go check out Droid AI
if you're listening.
You really should.
We'll get them on.
Good insights.
We'll get them on.
5, 6, 3.
It's mainly a self-hate.
Yeah, exactly.
Follow the chips.
Last segment. Last slide of the day.
NVIDIA dog feeding all of the startups in the AI ecosystem still.
Doesn't end.
They have an endless checkbook to continue to provide all of their potential future buyers
with more and more chips, more and more buyers with more and more chips more and more power
more and more funding every week i feel like nvidia writes another multi-billion dollar check
yeah i was gonna say i like just seeing the the billion like as we're desensitized to that stuff
now and a gigawatt of ruben gpus like that's insane i mean think about that your your unit of
denomination for a hardware product is in the wattage of energy it takes to power it
and a gigawatt is also the size like what what's it it's like a gigawatt it's like how much is that
again you know like your house runs on 15 likeawatts or something? No, no, no.
Not even.
Megawatts are, like, I think of megawatts as like city powers.
I think gigawatts are like an order of magnitude larger.
Maybe I'm off on my orders of magnitude.
But gigawatts are huge.
Like one gigawatt data centers are like, you know,
that's,
it's an enormous data center.
Yeah.
Power of the average American home.
It's 1.2 kilowatts.
Kilowatts.
That's right.
Kilowatts.
Our,
our metric viewers out there can do the calculations in their head.
Billions.
I'm sure we'll Williams does as well.
I don't know what you're talking about.
It's USA, man.
Come on.
Yeah, NVIDIA also did some kind of like saying
they're going to allocate a lot of funds to open source AI too.
I think it was in the billions as well,
which was I haven't dug into like exactly what they're saying,
but whatever, as long as there's
stuff running on their chips and people are buying
them they'll throw money at it
when does the ship stop
when the ships
stop maybe when the ships
actually stop because we don't have oil to ship
the chips nice
I think there's plenty of demand for these chips I don't have oil to ship the chips nice i think there's plenty of demand
for these chips i don't think that's much of a worry i think regulation stops it before the
demand does personally or power just like raw power i think it's more of just like that type
of constraint i don't think there's so much politics to get regulation. Like, I feel like there's enough money basically that would stop regulation from killing this.
That it more is like physical Adam constraints that we run into rather than social pressure.
Right. Well, that concludes today's show.
Thank you all for hopping on. It was super fun. I know it was very last minute and I really appreciate 563 and Seth joining me. But again, really exciting times. We've got a lot of weird, we're in the singularity right now, boys. Like I can feel it. I feel like there's so much happening. There's so many dollars floating around.
There's so much opportunity.
People are buying, you know, random agent social media sites for exorbitant amounts of money, one can assume.
We've got a completely alternative means of training and building AI with distributed systems that was not even once imaginable,
like two or three years ago. And it's a really exciting time. And I can really feel the momentum
that we're going through right now. But with that being said, thank you everyone for tuning in and
spending your Friday afternoon with us. I mean, that's absolutely awesome. I hope you guys learned something.
I hope you guys go and try those skills
and try those tools.
Go use that Cloudflare API as well
and just ask Claude,
what can I actually use this to build for my business?
And it'll come up with all kinds of different ways.
You are literally only limited
by your creativity and imagination when it comes to this stuff.
And to get inspired, you have to just go and execute and build things and talk to people.
And heck, maybe even listen to this show every once in a while and tap in and you'll hear something or get an idea that leads you down a rabbit hole.
But with that being said, again, thank you.
And make sure that you subscribe
to the Blockmates newsletter. Every Friday, we have the AI roll up hit and you can find all the
different links right there and you can read it all on your own time and get direct access to all
the tools. We compile all kinds of news and information and tools right there easily within
one email newsletter.
So thank you again and have a good weekend, guys.
Thanks, man.
Appreciate it.
