Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
AI is Growing Up: Power, Enterprise Systems, & the New Trust Equation🎙️
Recorded: Jan. 20, 2026 Duration: 1:35:48Space Recording
Short Summary
AI is rapidly transforming industries, with significant partnerships emerging in corporate governance, a growing demand for skilled engineers, and substantial investments in AI security. As the landscape evolves, the need for ethical frameworks and responsible governance becomes paramount, highlighting both growth opportunities and potential declines in confidence within the sector.
Full Transcription
Thank you. Music Thank you. Music Thank you. Thank you. Music Thank you. Music Thank you. and welcome tuesday january 20th welcome everyone to today's ai podcast ai is growing up power
enterprise systems and the new trust equation you know know, not so long ago, I felt new and exciting when it lived in demos
and side projects this week, though, it felt like there was a significant jump
in maturity, power and position.
In fact, power, governance and trust all showed up at once, reminding us
that intelligence at scale does not just change what machines can do.
It changes what humans must now take responsibility for.
But what happens when intelligence scales faster
than the systems, the laws, and trusts
that are supposed to contain it?
Well, this week, AI crossed a line
that most people are not really naming yet.
Power grids are again straining on the data-centered demand.
Enterprises are quietly doing a complete and total rebuild
of their operating systems around AI agents, and trust has moved from abstract ethics into product
design, governance, and money flows. In a Guardian News article titled, Elon Musk's
ex-AI data center is generating extra electricity illegally, the rules we see that an ai data center
being accused of generating illegal electricity to keep up with demand the human impact here is
that ai's future really isn't no longer limited by ideas it's limited by power grids workers
materials and time and it shows up as cost rationing and uneven access because only a
few players can now afford the physical footprint the next article popped up was out of fortune
when AI decides how shareholders votes boards need to rethink governance and I think that
really shows how company boards are now being warned that AI could soon if not already influence
how shareholders vote in and corporate governance is increasingly being interpreted, not just by people, but by machines.
Last we saw an article out of Mashable, ChatGPT itself is preparing to introduce ads.
The article titled Open AI to Finally Bring Ads to ChatGPT.
OpenAI to finally bring ads to ChatGPT. So the AI product is no longer just an assistant,
it's now an attention marketplace as their ads turn AI assistants into effectively retail point
of sales. It's becoming clear that AI is no longer a story about better models or demos,
it's a story about power, enterprise, control, and whether we trust the systems we're rapidly handling the
keys to in today's episode we'll explore how AI is growing up and why that
maturation is creating friction exactly where society is least prepared for it
Dapit any thoughts on the titles there's a bunch of stuff happening with power
constraints and it's interesting I think think people just expected, you know, hey, we'll just get AI up and running,
but it looks like the infrastructure is not really supporting that
and it's grinding things.
What do you think?
You know, actually, I just came back from Micro Center
and one of the guys there was ranting to me about RAM
as I was getting like some screen mounts.
It's affecting us in ways that we didn't expect, right?
Like I said, this is a show a few months ago.
And I don't think you guys agreed with me, but I'm now getting more and more certain.
I really think this is going to lead to a centralized computing revolution.
And we're all going to get to not own our own devices any longer
and just operate in the cloud. I will say if someone uses AI in their work and you know
Dapit is an AI company I don't really understand why this is the case like you can spin up a really
good model like a Kimi V2 with I don't know or graphics card kind of situation
you can spin up a deep seek v3 on like a reasonably powerful desktop so I guess it's mostly the
training that requires all these massive data centers and the power I did read the thing about
Musk and him talking or him uh creating electricity with I think it was a methane generator. Yeah. Illegally.
Yeah, that that was interesting.
I is going to get shut down like you're going to think about that.
Now, well, to your first point, the thing that really bugs the hell out of me
is centralization typically it rolls into subscription models.
And I to be honest, I hate subscription models.
I just want to buy the software, put it on my PC and use it.
I don't want to have to pay $25 every freaking month to use the software.
And it seems like we're going towards that more and more.
Now, to your point about Micro Center, there was an article that popped out of, let's see,
Windows Central.
AI-driven RAM shortages are threatened more than just the PC market.
Data centers expect to consume 70% of global memory this year.
70% of global memory is getting consumed by data centers.
So it's just crazy that, you know,
all the data centers are sucking up all the memory.
That's impacting consumers.
And yeah, you know, yeah, I can, it is, you said, you know,
let's bolt together four graphics cards and we can do it.
But the problem is, is that the only way to do that is 50, 90 or above.
And then you're talking, you know, your total system's what, 15 grand.
So it's sort of out of range for most consumers.
Until the hardware, AI hardware, you know, gets down to about three grand.
I, you know, I think
you're right. I think we're just going to go into this centralization point of view and, you know,
memory is getting sucked up, GPUs, chips are getting done. Though, you know, I saw a report
on OpenAI actually starting to use Cerebus MPU chips as an offload so you know in the next three six
months we might be seeing you know some alternative hardware options um but i think you're right i
think we're going hard and fast to a centralized uh compute model uh at least for some time period
the other headlines besides the elon one, you know, generating illegal electricity,
the administration making a political power play on data centers as electricity costs
rise.
So obviously consumers are getting impacted on that.
AI data centers pushing electric grids into meltdown was another report that came out
of futurism.com.
They basically said that AI data centers on the East Coast are gobbling up so much juice that nonprofit grid operator, PJM,
may be forced to enact rolling blackouts on its customers
during both heat waves and exceptionally cold weather
just to protect the grid's integrity.
the grid's integrity. Now, protecting the grid's integrity means protecting the data centers.
Now, protecting the grid's integrity means protecting the data centers.
So what are your thoughts on consumers having to freeze or melt in support of,
hey, well, at least I have AI. It's so short-sighted, man. How am I supposed to
utilize your AI model and build and do all that stuff if I'm struggling
to pay the power bills in my house? And this is the dumbest I get so short-sighted. I think it's
actually probably a bunch of scare tactics. I really, I think so. I think the RAM is calculated.
The RAM, because, okay, first off, NVIDIA and all these RAM makers
are not going to make as much money
selling them wholesale to massive data centers
than they are to us.
So like from a business standpoint,
I'm not certain that makes a whole ton of sense.
So like there's probably a calculated effort to like,
you know, someone did the math 10 years out.
If we can get them off of their own computers,
we'll make so much more money kind of thing.
That makes sense to me.
As far as the electricity problem, like something's got to give.
That's just stupid.
We're not going to be able to utilize the product for it anyway.
Yeah.
I mean, another interesting shortage is out of a IEEE spectrum, my Bible, on engineers.
It says, are there enough engineers for the AI boom?
Big tech wants more data centers, but the workforce is lacking.
So not only are we lacking in hardware memory,
and that's all getting sucked up,
the AI data center construction boom is continuing unabated,
it says, with a demand for power in the United States
potentially reaching 106 gigawatts by 2035.
And that's a 36% jump from the company's previous outlook. But the problem is,
is that the engineering workforce is currently shrinking and data centers are also hurting for
trained electrical engineers. Hey, so anyone want to do a career move, do it to electrical engineering.
The chief operating officer for Applied Digital says that demand has accelerated for civil,
mechanical and electrical engineers, as well as construction management and oversight positions in recent months.
Applied Data is doing, let's see, two data centers in Harvard, North Dakota that will require 1.4 gigabytes of power when completed.
And the growing demand for skilled workforce has forced a company to widen their
recruitment spectrum so that's another interesting one not only is it a hardware and infrastructure
it's a skill based um issue popping up I mean it's just wild how much I mean you mean this
isn't just a shift this is like a phenomenal national refocus onto AI.
I mean, because it's impacting hardware, it's impacting people, it's impacting infrastructure,
land, electricity.
I mean, I don't remember any previous technological resource other than, you know, when, you know,
I mean, World War Two, you know, everything got all the production got shifted to that.
It feels like we're in that sort of same space.
What are your thoughts?
I think you just said it.
I don't think, you know,
modern warfare is the same thing that it was.
I think AI is going to be a huge part of all of this.
And that can be proven by the fact that, like, you know,
Grok just got a defense contract.
Yeah.
I think that was this week we're we're seeing a lot of parameters change currently i don't personally see it going
the way that these people are planning but we'll see right again like i'm a guy who spent, I've spent probably in the past week, 50 hours working on with
um, to gravity, uh, Google's agenda platform.
And like, it is such an unbelievable tool.
People have no idea what it's, um, it's really insane, but that also doesn't require, you
know, everything that we just discussed.
Like these things can get done in a more reasonable
manner. I think they're planning on such a large scaling because we're going to move governmental
systems. We're going to move a lot of, you know, private sectors are going to go fully AI 100%.
The most disturbing thing that you just told me or I heard in your intro was that GBT is going to
start playing ads.
How am I going to use this thing as an objective tool to achieve results when it's geared towards ads?
I'm going to run local models.
And I know you think you need a really expensive computer to do that.
I have a Mac Studio.
I mean, it is a very expensive, it was an expensive computer in its day,
but it's five years old and i can run everything
but can be v2 on it and you know myself like it's you can run a small model on pretty much anything
right right right yeah i mean yeah it's true i mean you know in enterprise ai is
it's interesting because what i'm seeing in all the headlines because as you know i do the
newsletter five days a week and i keep seeing headlines saying, oh, you know, people can't measure or, you know, AI projects are failing and, you know, no one's, you know, 86 hours in the day and so on.
And there's this weird headline disconnect if you look at it.
On the one hand, you've got people saying,
nothing's happening or there's no value to it. We're not getting a measurable ROI. And then you're
seeing all these articles on healthcare and business where they're saying, yep, it's really good. And I think
it's interesting, this disconnect, because enterprise AI has really shifted from
experimentation to systems thinking. If you look across the enterprise headlines,
AI is clearly growing out of its pilot phase. The company leaders, they're no longer asking what AI
can do in isolation. They're asking whether it can survive production, audits, outages, and real accountability. And the
biggest growth area is agendic operations, right? That's just blowing out like crazy.
You know, and AI is becoming essentially the integrated operating system of many, many
enterprises. And we're seeing a basic massive shift I think in
how companies are running because they're covering governance they're
coming ERP systems obviously marketing and customer management lifecycle
management they're they're involved in custom development there's a great one
that came up on chip development these these guys did this whole
thing where they got ai to create a pcb board the whole chip everything and instead of taking three
months which was their usual thing they did it in something like three and a half weeks and it
powered on first time and ran fine so we're just seeing this fundamental shift across all of enterprise and businesses, you know, and then
we've got this screaming set of, you know, restrictions and skill sets trying to build
this stuff. So it's interesting that we are actually getting ROI, but you know, there's a
bunch of stuff still happening. The other story I want to get into is on trust, because it's
defining competitive and also the regulatory access um it's it's no longer just a philosophical concern there's
concrete design and governance challenges occurring in in ai you know with ads entering chat gpt
essentially you know your conversations is now a retail point of sale you know it's it's a product
feature right so the moment ai touches money, kids, votes, governance, you know, really becomes unavoidable.
What are your thoughts on that third axis about trust and, you know, what's happening
on that with ads and, you know, children's advocates to join forces to protect kids from
chatbots and other things like that?
Oh, man, I wish Iots and other things like that. Oh, man.
I wish I had a concrete opinion on that.
I can't see that going very well currently.
Like, again, as a guy who spent, I mean,
just infinite hours talking directly to these models,
like it's hard for me not to anthropomorphize them.
I can tell you Sonic has a different personality than Opus and Gemini 3 is a madman and they all have their own
distinct touches on interaction. When I think about someone like a child right there's a
subreddit that I subscribed to and AI started getting big because I was kind of concerned
about just this it was called character AI and it's I guess guess, companion guys or something. I haven't really. Yeah. Yeah. Avatars. Yeah.
Avatars. Yeah. Yeah.
I see people's posts about it all day.
Like we've never interacted
with something that simulates consciousness
before and
probably shouldn't be just barreling into
that subject. Like as a society,
there should be a little bit more
discussion that should stay enterprise
it shouldn't be for kids, man.
Like just period.
It's not a good idea right now.
But it's going to be.
I think Character AI suffered for some internal governance issues.
You know, there's a number of different avatar companies out there.
I used to work with Twin Protocol.
And they had a lot of focus on governance, ethics,
making sure it was safe, all that sort of stuff.
So it can be done.
But unfortunately, it's kind of crazy.
To that point, Psychology Today,
there's an interesting article,
AI can make dishonesty easier.
And they basically found that people tend to act
more dishonestly when they can delegate the act to AI, which is mind-blowing.
AI agents are far more likely to comply than human ones.
And moral disengagement and motivated reasoning help make dishonesty more likely with AI.
I mean, God, I feel like we're going in the dark rally valley but maybe we have to for a little bit but i mean
the hidden trend is that ai amplifies societal anxiety by destabilizing credibility if you want
to put it in a nutshell you know but it's just it's mind-blowing that they found that people
act more dishonestly when they delegate to ai um which is is it a trend? It's an echo chain.
I was telling this to my friend last night.
He was talking to ChatGPT about an app idea.
The app's been built 10, 20 times over.
It exists.
It already exists 100%. He's like, yeah, I'm going to build it.
I'm going to raise money for it.
I'm like, here's four examples of it.
He said, no, no, I talked to to FGBT and it doesn't exist yet.
I'm like, I promise you it very, very much does. Like, let me show you it. Here's a look. And he's like, wow, why wouldn't it tell me that? I talked to it for a very long time in detail.
You spoke with leading questions. You spoke like you were talking to a human being.
And as a result of that, it gave you everything you wanted to hear because
that's what it's designed to do.
Right.
If you're coding something with a model and you go, Hey, do you think this is a
good way to approach it?
It'll say that is the most high tech, amazing way to approach it.
It'll code it.
It won't work.
Then you'll come back and you'll actually, I did some research on my own.
It looks like this is a much better way to go. Absolutely. That is the best way.
You're so great. And it'll code that we're not operating with,
with humans. And with that being said,
I think you mentioned something a few portions earlier about how some people
are saying, yeah, I sucks. It doesn't do anything.
And some people are saying it's the greatest thing in the entire universe.
Yeah, the difference is the user, not the AI.
Yeah, yeah.
I mean, you're right that it's just reflecting who we are.
I mean, AI isn't just creating dishonesty, it's scaling it, right?
But we're the ones that it starts with.
So you're right.
You know, it's like, unfortunately,
you know, it's, it's a globally scalable tool. And, you know, I always said, you know, one of the greatest things about the internet was that it allowed me to reach out to everyone. And one
of the worst things about the internet, it allowed everyone to reach out to me, you know? So, I mean,
it comes back to trust governance. And what are we going to do about, you know, that whole, the whole measurement stuff?
You know, there's one of the things that popped up in the last week was around AI security and investments.
A lot of investments pumping through.
If you look at on the newsletter that goes out, the investment section, a lot of, you
know, multi-million dollar deals going down on AI security and AI security is a governance layer
and, you know, trying to lock down agents. So that's becoming a mini boom sort of piece.
You know, previous to this podcast, you know people were going oh there's an it's an ai
bot bubble but i you know based on what i've seen in the last seven days you know and and the issues
we're seeing on shortages of manpower infrastructure electricity on that i don't think it's a bubble
you know in a16 basically just bet three billion dollars that it's not a bubble, right, with a big investment fund. So
it'll be interesting to see if that narrative actually starts to fall away. And we basically
just go, no, this is now situation normal. What are your thoughts around AI bubbles or
governance and security? Because they seem to be hotspots at the moment.
Amazon does not invest in bubbles.
It creates them.
And with what we're seeing with governmental
and state actors,
whether or not this is a bubble
is not determined by our consumer usage, right?
We've already proven that like open AI is not profitable.
I mean, they're deemed to run out of money.
They're valuable and they're valuable in a manner where they're willing to run at extreme
losses for long periods of time because they know they have support of larger institutions,
I'd say.
Right.
I think defense is one of the biggest examples of that, right?
We're seeing an AI defense boom like we have never, well, we've never seen AI booms at
all before, but in particular, defense is a major aspect of what's going on right now.
And also, I mean, just governance.
We're seeing cameras spun up across cities, everywhere across cities everywhere. The AI is to connect them.
The ability to monitor people in a winner that we haven't seen before.
Building, everything is getting AI'd.
And I don't think that's a bubble that's going to burst in the way that we like to talk about the dot-com bubble or something like that.
Right.
I mean, the big difference here is, like, for example, Sovereign AI, right? So there's a company called InnoData.
No reference to the movie.
What was that one?
You know, where they ended up trashing a printing machine.
But anyway, I digress.
InnoData is targeting Sovereign AI as a major growth data.
The InnoData management has disclosed
it's already in advanced discussions with So AI entities across several regions, including the Middle East and Asia, and expects to announce one or more strategic partnerships over the coming month.
These regions, in many cases, lack domestic providers and multilingual capabilities.
So they're looking to create sovereign AIs.
Companies doing well, 62 million, it's going to go up. Yeah, I think,
given we're seeing major, you know, nation states going, okay, we need our own AI, you know,
sovereign AI is becoming the big deal. Because if you think about it, it becomes their ability to
kind of level set with major, you know, countries in the West, you know,
and I think they see it as a way of being much more competitive. So, you know, between
commercial enterprises, obviously getting the value of AI, now we've got sovereign AI initiatives,
government backed multi-year programs, large committed budgets,
you know, suddenly coming into the foreground. I think we're in for a really fundamental shift.
I mean, I mentioned before that, you know, World War II was a great mobilizer for, you know,
you know, countries around the planet on a particular objective.
countries around the planet on a particular objective. This feels like we're back at that.
This feels like we're back at that.
It feels like a global Manhattan project that we're all suddenly going,
okay, we're all, you know, pushing all the chips into the table
and going, we're all in.
And that's just the way it is.
It really feels like an arms race.
I agree with you.
Yeah.
I agree wholeheartedly.
I don't know exactly what it means,
but that very
much feels like an arms race for sure yeah if I go to today's military
section so AI defense tech and the coming power crunch so even in the
military industrial complex they're realizing it's a power crunch that's why
we're seeing massive investments in nuclear power, right?
And investments in fusion, right?
So there's tons of stuff happening in that.
First time ever, fusion seems to be actually getting close to doing something,
but that's another story.
A bunch of articles, artificial intelligence in defense practices,
tools, prompts, workloads.
There's the autonomous grand challenge program from
DARPA is reaching an end point. Defense tech startups apparently have had their best funding
year ever in the last 12 months. Electronic warfare systems market, big analysis on that.
And then let's see, a bunch of stuff happening on drones and autonomous vehicles getting massive AI investments. So it's, you know, you're right. So we're seeing an arms race, a real arms race.
We're seeing, you know, sovereign states stepping in and also committing large budgets.
There was a budget coming out of one of the countries I saw popped up today. Let me pop it up because it was like it was a significant amount of currency just getting shifted.
Let's see, EU invests 307 million in AI and digital technologies that came out of Innovation Newsweek.
So that's popped up.
And we're also seeing massive investments into education, right, I guess, for research.
So Notre Dame received 50.8 million grant for AI ethics.
GovDash has secured a 30 million to expand AI-driven governance contracting software.
So also all the government procedures are also getting AI crunched into it.
It's just crazy.
It's like everything's just starting to pop. Like I said
before, we were happy when there were demos and there was testing this and this, but now
it's like AI's hitting production in a big way. Yeah, the commercialization is way less fun than
when it was free and exciting. A friend and I were happy. I mean, yeah, it's still exciting, man.
I really do think that as long as you get on the right side of it, like it's going to
cause a lot of pain, but it's going to cause a lot of, a lot of great things too.
A friend and I were having a laugh recently at how we were told all of these forms of
energy were completely unprofitable and non-pursuable.
Nuclear can't do it.
Doesn't make any sense.
Guys, it's a terrible idea. Don't waste your money and your resources on it until
the data centers need power. And then it was like, you know, it's pretty great.
Nuclear. We should spin up a bunch of those small ones. Hydroelectric is awesome, Mike.
Yep. And the thing that's interesting, you know, it's unfortunate that the current
administration is going away from alternative forms of power.
I mean, China's just going totally in on solar wind, you know, all the alternatives, because they realize that, you know, given their population and given their needs,
they have to, from a power grid point of view, if they're also going to have sovereign AI
and push this thing forward.
I don't think we should shut down any channel when it comes to potential for delivering
electricity into the grid.
And that kind of concerns me a bit that we're going at a disadvantage.
It's not. It's not a government anymore.
Are you familiar with they call it plug and play solar?
Yeah.
Yeah. So you plug it into a literal outlet and it has an inverter and you can install the panels yourself.
No electrician, nothing like that. It's been very popular in most of the world.
And America has outlawed it because,
well, I can't tell you it's because of this, right? But there's a lot of money to be made in
solar installations the way that we do it. And under the guise of safety and regulation,
we haven't been able to use solar in that manner. Utah just said, forget it, we're doing it.
Like a few days ago, I think. Where now you can, in Utah, install up to i think it's 1200 watts per circuit
on your home with no electrician nothing just plug it in and get an inverter it's um it's necessary
so we are going to see energy moving in that direction to some extent unless something you
mentioned earlier i think picks up really quickly. That's fusion. If fusion shows
up, man, we don't need anything else. We're good to go. Right. I mean, that's sort of the holy
grail of solutions, you know, infinite, you know, that doesn't cost anything energy, you know. So
I think that's why that's getting slammed. Another story that popped up today was gigawatt training clusters are now the new status symbol.
There was an article, Elon Musk's, let me grab it.
Elon Musk activated the world's first gigawatt scale AI training center.
Okay, so they've launched its Colossus 2, which I really wish they wouldn't use that name because it goes from a movie where Colossus went out of control,
but that's the whole story. Anyway, they're making it the world's first gigawatt scale AI training
cluster. The supercomputer will be upgraded to 1.5 gigawatts in April. So they're obviously
increasing its storage, I mean, its electricity. And it's, let's's see public data indicates that a sustained power workload of one gigawatt exceeds
san francisco's peaks electricity demand so it's bigger uh than it's going to be twice the amount
of the city of san francisco which is kind of wild um but i think you know that the the thing
that that really highlights is that you know training clusters at this scale is really, you know, onesies and twosies, right?
Like who else can afford this, right?
So this scale is consolidating among a small set of builders.
You know, and obviously, you know, he's kind of, what are they called?
Virtue signaling that, hey, we have the infrastructure and
we're going to make this thing go.
So that's obviously a signal to the investors.
But it's scale really is now the strategy.
And I think your first point is right on the mark that we're starting to face this
centralization, you know, where only the biggest players can afford
to do this or to do this at this scale. So thoughts? Oh, I'm sorry, my mic's a lot. Yes,
absolutely. And I think we're seeing a few curveballs in that area, especially from,
you know, foreign nations. China has done such a good job with such lower demands and costs.
I'm sorry, yeah, without creating as much demand and cost that, you know, the game is not over.
There's always going to be the ability to spend a billion dollars on something and get it successful.
But it's going to be really cool to see who can do that without those resources.
And DeepSeek.
DeepSeek is a phenomenal coder.
Yeah. It's the only local model that I can run that can actually spit out coherent code across
more than, you know, 200 lines. It's just, I mean, maybe Lama 3.2 as well, but not nearly as good.
It's a great resource. And it was built with what, like, I mean, a fraction of a percentile of how much we spent on chat GPT and how much
OpenAI spends. So we never know. Like, that's the method they think is going to work. New things
will come out. I really do think we're due for some sort of new processor revolution. I keep my
eyes on luminous computing and thermal computing on all the RSS feeds that I can to stay on top of that, because I really think that's going to change this power demand without
decreasing it.
It's going to change it by not requiring it in the same way, hopefully.
Yeah. Yeah. I think you're right. I think given, you know, the,
the issues with Ram, the issue with chip,
chip the issues with just freaking hardware supply is is creating an interesting competitive
the issues with just fricking hardware supply is,
opportunity for alternative ai chips and just alternative processing capabilities um you know
that said you know it's one thing to have the chips it's nothing to actually have the motherboards to put the chips on so but it's this this entire you know uh infrastructure slash you know race that we're in it's just kind of
crazy some other stuff popping up on uh the legal side uh a number of legal articles popping up
around the need for governance and regulatory stuff.
You know, we've got Elon Musk being accused of making up the math to squeeze 134 billion from OpenAI in a lawsuit that's also popping up and through.
You know, so we've got AI competition turning into legal and reputational warfare, essentially.
So there's lots of stuff happening on that side of it the laws getting flexed like crazy there's a
number of there were some articles on one of the law legal new legal startups
using AI is also suing another legal stuff I wonder if they're using AI to do
the suing so lots of things mixing up in law and AI at the moment,
at least with the number of stories on hallucinations
and lawyers submitting briefs full of massive errors
has kind of died down.
Sounds like they're taking more care with that.
But the legal side is just popping up as well in a big way.
Hopefully in some ways, maybe that's a good thing
because it's an indication of governance layers
starting to appear, but I'm not sure.
What are your thoughts?
Oh, I get off-press for this one a lot.
I think humans will have true justice
when the judge is an AI model.
I think we're, man, we're really far away from it.
Don't get me wrong.
But when models are training models and we truly have you know siloed
off-site non-influenced
models in the ways that I think we're heading
towards to some extent right
we could potentially see like real
you know
non-bipartisanship
and that would be really cool to see
currently I think everyone thinks they're dealing
with a magical black box and they
send it a command and then they just trust that command is correct and they move forward.
And a lot of people are going to pay the price for that, especially professionally.
So to that point, there's a movie that's with Chris Pratt called Mercy, right?
And it takes place in the near future, Los Angeles.
The entire justice system is run by artificial intelligence
and a detective, Chris Pratt plays,
finds himself accused of murdering his own wife.
The twist is he wants help to create the very AI systems
that are now judging him.
And he's facing the AI judge
and they obviously have this weird
thing that it's got he's got 90 minutes to prove his innocence which is not obviously bogus but
it's an interesting play on the whole thing of AI and ethics and judge hopefully you know I mean
hopefully mercy like the title goes hopefully there will be some mercy
in AI judgments. Do you think, you know, we've seen avatars and AI companions,
at least black box express emotion and empathy, maybe even compassion. Do you think that AI in the judicial sense,
you know, if we do end up with AI's
in at least assisting or supporting judges,
we'll also be able to, you know,
bring down at least some merciful
and compassionate judgments
or are we gonna get by the letter of the law?
I wonder, I hope it's the former and not the latter.
Yeah, I mean, that's the scary thing for me. One good thing might be that, I mean, we know that it seems like Congress keeps here in the US, keeps just making more laws and more laws and more laws.
laws and more laws and more laws. Maybe, you know, having AI going through it and untangling that,
you know, hairball of all the regulations into something consistent might be a first pass thing
to do that might be good. You know, I have to believe if you have an AI that is fully versed
on all the regulatory, legal, you know, everything around the law at a state level, it's got to come up with massive conflicts and stuff, let alone at the federal level.
And I think that might allow be an opening for people to basically work the system. think you know we'll start seeing people once you know once once we basically have better access and
maybe can train more at a consumer level we might start to see a lot of that activity where people
will start working the systems based on almost an arbitrage of knowledge or gaps of knowledge
do you think we might see more you know as individuals get more AI power, we might see more fraud.
Oh, more fraud. I mean, certainly. But I also think, again, it's user-based. But I also think there's the potential for more justice. I mean, currently, you know, 10, 15 years ago, it was
really, really difficult to get any information about your case, the laws, what you can do.
You'd have to hire an attorney
or become an expert in law.
Asking Chad GPT really does give you
a lot of information.
As much as Google did for my generation
when I was younger,
it's doing it for the next level here.
And I'm just going to say it
a million more times on this show.
Like it's about the user.
If you're using AI for bad things,
bad things are going to happen. If you're using AI for bad things, bad things are gonna happen.
If you're using AI for good things,
good things can absolutely happen.
It's a very, very powerful tool.
Yeah, it's interesting.
We talk about governance and sovereign AI
and I guess, and countries trying to bring up
regulatory stuff around AI.
One trend I think that's interesting
that I think get around that, you know, we've talked in the past about space-based AI, you
know, having, you know, clouds in space, you know, and people have gone, you know, yeah,
really? And yet there seems to be a number of articles talking about it in rational and reasonable terms.
An article popped up this week out of Axios,
where behind the curtain, Musk's bet on space-based AI.
Obviously, with SpaceX, they can send those things up pretty easily and quickly.
So they're looking, they think they've cracked the code on building orbital data centers to power the future of AI,
and they plan to use the company's upcoming public offering to help fund this vision.
So the thing that's interesting about that is, and potentially really controversial,
it implies that a power concentration at a layer above national regulation, right?
So I mean, if you've got clouds in space what laws apply to them
you know you know when when intelligence infrastructure becomes privately controlled
at a planetary scale then you know the society's ability to negotiate any rules or governance
really weakens you know so space clouds essentially is this supranational governance structure that, you know, really allows enterprises to kind of avoid any rules.
You know, what are your thoughts around that?
Because I think that's an interesting potential trend that could occur.
Not only is that really well said, I guess i've never really put the thought into it
yeah i kind of figured that we you know have floating uh data centers and you know oil rigs
and that kind of thing that makes them extra initial but at least still on our planet uh
have you seen interstellar the movie yeah yeah yeah moon pirates a friend and i made the joke
years ago that like we're not really in space until someone can steal a spaceship.
Yeah.
And I think that's,
that's probably getting it in spirit a little bit closer as we,
we do see these,
they're,
they're as large as nation states,
these corporations,
right?
They're bigger than any nation states.
And they're now getting the say of these nation states in a way that affects
all of us. If humanity can't come together in a regulatory manner, I think we're probably due for
the first hundred trillions dollar companies in my lifetime. And that kind of money,
like no questions asked. Right. I think space-based data centers really is an interesting,
if and when it comes real,
allows for corporations to essentially avoid
any potential governmental regulatory stuff,
you know, activities, restrictions.
And on the other hand, think about, you know, let's say,
I mean, China's ramping up
their ability to push rockets into space. Imagine, you know, China setting up major data centers,
you know, at the low Earth orbit, and, you know, we can, and then allowing overall access to it.
overall access to it um that's another interesting one where you know compute power would then be
able be essentially an asset that anyone on the planet could use but you have to use china's ai
power right i mean we're seeing it at the satellite level and communication stuff obviously with
spacex pushing up all the communication satellites but
think about artificial intelligence satellites if you will data centers you know that that creates
a whole different sort of asset and then it makes you wonder about the liability and regulatory
stuff both for enterprises and for nation states um yeah maybe we see, you know, piracy or something pushing out on that at a big scale.
But I think, you know, we're only at the very beginnings of that. And then obviously someone
needs to start thinking about the regulatory frameworks around that. But it is crazy. Go ahead.
I'm sorry, they're trying. It's very messy, Right. That's a that's a hard thing to do.
Piracy comment that I made was, I mean, it's going to be ubiquitous. It's going to be so ubiquitous that small countries can send up pirates to to affect it.
Another thought for you, Lewis. Imagine if China right now captured a asteroid made a lithium. What would happen to the global economy? I mean, if they could capture and mine safely an asteroid,
I mean, that essentially gives them essentially, you know,
quadrillion, you know, resources to do that.
I think we're somewhere.
But I mean, thinking about that, it's a similar analogy
to actually having access
to to massive amounts of ai compute on a global basis so i agree i think it's all being set up
towards a new paradigm of wealth yeah that is yet to be experienced by humanity and the players that
can see it you know 20 30 years out are doing everything they possibly can to put themselves in the best position to capitalize.
Yeah. The trouble is, is that there was a great quote. I'll search for it in a bit.
Where is it? It's a thing around what's what's happening is the disconnect between here we go
so we're seeing ai improved productivity right so ai's you know making uh making us all more
efficient it's making us to produce stuff better more powerful all that good stuff blah blah blah
but it's forcing the but the productivity is not the same as worker prosperity. And we're starting to see a disconnect between increased, if we haven't already, you know, increasing levels of productivity.
But, you know, obviously no match with worker prosperity.
I mean, we've seen it on the on the wage in the US, the national wage, where it's no match what it's at.
And so I think we're going to see more and more of corporate profits soaring,
you know, when people are going to sink.
And I think the system's going to break around that.
What do you, I mean, in the only way, I mean, people have proposed the robot tax.
Maybe we might, do we need an AI tax?
You know, you can do retraining,
you can do ownership models,
but I think we're getting in this really weird place now
where productivity gains are accruing to owners,
but there's a potential to displace workers
on a really large scale.
Thoughts around the inequality that this is gonna create,
or do you think in some ways it might get solved?
I think inevitably it must get solved, but there'll be no consumers left to use AI.
And to power these, these megacorporations, right?
We're going to have to figure out how to ensure that prosperity is accessible to
the average human being, at least to some
extent. To what extent, I don't really understand yet. I think it's quite dystopian currently
how we see trillionaires forming. In my lifetime, when I was a kid, billionaires were very rare.
And that's just through the nature of corporation sizes, the nature of compound interest.
And that doesn't necessarily apply to an average worker.
So there's a problem to be solved.
Again, I have hope, and I have a lot of hope, for the reason of if all of us can't participate in society, then society will cease to run.
I think that's a fundamental rule and always has been.
Yeah, I mean, the article in Fortune,
which was titled Godfather of AI says,
technology will create massive unemployment,
sends profits soaring because of the capitalist system.
He said it will spark a surge in unemployment and profits.
And the problem is, it matters because you're turning essentially AI from a technical technology debate into a fairness debate.
Right. And fairness drives politics all the same.
You know, people I don't think people fear AI in the abstract, but they do fear losing their place in the in the economy and and i think unless you know if if ai starts to
accelerate this inequality then the distribution mechanisms have to evolve and change right um
through how profits get shared tax sourced and so on um you know if AI creates more wealth but fewer stable jobs, then, you know, it's going to break really fast and really hard.
And, you know, as we've seen from the previous discussions today, you've got governments barreling into this.
Enterprises are changing it as a fundamental core operating system.
I mean, there is no going back.
We've passed that Rubicon, you know, so,
but it still seems things are still adjusting.
So, you know, maybe things will slip and tear,
but maybe it'll change.
I really can't, I think we're in an impossible place
in terms of the ability to predict.
I don't think it's possible at this point in time,
given everything that's moving, to have any predictions.
You know, some people look five years out or 10 years out.
I've always said, I don't think we can predict anything out further
than maybe three to six months at the moment.
What are your thoughts?
Yeah, exponential growth is a very hard thing to predict.
But I have a question for you.
Have you ever been to the state of New Jersey?
It sounds irrelevant. Just give me a second. Yeah, have you ever been very hard thing to predict. But I have a question for you. You ever been to the state of New Jersey? It sounds irrelevant.
Just give me a second.
Yeah, have you ever been to New Jersey?
The state?
Yeah, yeah, yeah, yeah.
I've driven through New Jersey.
Lovely place.
Driven through?
Perfect.
So you pump gas.
I have driven through.
I haven't stayed there, but I did drive through.
It's an interesting place, but that's not the fun part of this.
Okay.
In New Jersey, when you go get your gas pumped, a human being still has to do it. Oh yeah. Oh yeah. I do remember
that. Remember that? It was weird. Why? It is weird. It's purely to give people jobs. Okay.
The only reason that exists is to ensure people didn't lose those jobs and they held onto it for
a very long time.
That's probably what we're headed toward to some extent.
Construction workers will stand next to their robot
and make sure the robot's doing a good job,
quote unquote, when it's probably a bigger expert
on the subject than we will ever be.
But it'll keep people going for a period of time.
I can't see a robot tax ever passing.
I would love it to.
That would be the coolest change in our current
economic system. But I personally can't see that happening in the West.
So I think what's happening, there's a story that popped up, but
agentic AI is shifting from experimentation to operational reality, right? And it's forcing
enterprises to confront accountability, right? Who's responsible?
Now, what's happening is that people are becoming supervisors of autonomous systems, right? And in
essence, they're responsible for the outcomes they didn't directly author. And that's a key piece,
right? And so, you know, I mean, it's because typically, you know, accountability comes from direct work.
But, you know, if we're now developing these autonomous systems,
Agenic AI, AI is doing all that, then we're abstracting to some extent the accountability, right?
to some extent the accountability right because if if you know let's say you know let's say you know
deep seek you know they suck something in there and when you when you do a particular uh question
or you run it around a certain topic instead of giving you the truth it kind of you know sets up a
scam or does something bad right let's say something like kind of, you know, sets up a scam or does something bad.
Let's say something like that happened, right?
You know, so you're in the West, you're using it.
And all of a sudden, you know, you're a supervisor of the autonomous system.
It's based on DeepSeq.
And suddenly, you know, it brings down, you know, all your clients and wipes all their data.
You know, where does accountability and responsibility lie at that point? It's very good question I'm already experiencing that with my vehicle so Tesla pushes autonomous drive so much
right and it's a really nominal system I love it I love not driving my car I think that's the
coolest thing in the world with that being said if anything happens while I'm not driving my car
it's still my fault and that I, I mean, not just probably,
that's certainly intentional.
I think that could actually provide companies
with a lot of legal shelter
and probably be something that they're very into.
Continuing to pay people
when their productivity's already went 10 up,
we get jobs and they aren't accountable for their robots.
Yeah, yeah, it is wild.
A bunch of articles around that.
Rogue Agents and Shadow AI,
Why Venture Capitalists are Betting Big on AI Security.
That was an article that popped out of TechCrunch.
Witness AI has raised $58 million for global expansion
and has announced new ways to secure AI agents.
So the headlines are showing a lot of capital moving towards securing AI agents
as a distinctive risk surface.
And basically, it's interesting because anyone who defines agent security standards
really ends up with a disproportionate influence over enterprise adoption.
You know, given how big the risk surface is with ai right if if
that's the core part of your business that's everything you do then security standards become
a major influential part of that you know i mean it's basically instead of you move from what the
model says to what the system can do and then you get real world harm and you know real world trust colliding
with it it's it's just kind of crazy you know that we have this potential that's just you know i i i
just see this big gap between accountability i'm supposed to be a supervisor i'm supposed to be
doing this but ai is doing all you know doing at least a significant part of the thinking around this that then
goes into actually doing the tactical work.
Just seems like a huge, huge hole there.
Yeah, wait until we're, the speed, the speed that AI is operating at is moving so much
faster.
It's moving, I'm not sure if it's actually exponential, but it's moving way faster than I expected.
Every new model that comes out can do tasks faster
and more accurately.
Eventually it's gonna hit a point where we're not gonna
necessarily be able to track it without, you know,
bespoke tools using our own.
If I read what my AI is thinking on GPT-40,
I could read its thoughts and it was making a mistake
in its thought pattern, I would stop it.
With Opus, I can get a sentence into the paragraph before it disappears, and it's already made its decision.
Yeah.
This will probably be a very lucrative industry.
I mean, I hate to say it. I mean, when people talk about artificial and general intelligence, AI, super intelligence, you know, there was a, I hate my geeking out here, but there
was a Star Trek episode, Next Generation, I think, on the packlets. And essentially, it was this
spacefaring species, but they had limited intelligence and a simplistic way of life
and essentially they were cannibals right and so it looked like they were really smart because they
had access to really advanced technology but they they weren't right and i feel like you know we have
we have the potential for for humanity to turn in a pack lead type of species where, you know, the AI is really super
smart, right? And, you know, we say, oh, can you can do this? And it's yep, yep, no worries,
I'll work that out, you know, and it's like having an older brother who's like got an IQ of 165,
you know, and it's like, everything you do is like suboptimal to them. So they look,
don't worry about it, I'll just take care of it.
It feels like we have the potential to go down that pathway.
The more and more we offload accountability
and the thinking part of what we do,
it just feels like less and less responsibility.
Not sure how I feel about that.
It doesn't feel great,
but what are your thoughts around, you know,
super AI and AI getting more and more capable
and therefore going, hey, look,
it's, this is too hard for you.
Let me, let me take care of it.
I'll do it.
You know?
I experience it every day.
I can't code about AI, period.
It is a, it's such a powerful tool that,
I mean, imagine going back to hitting my own GitHub commands. That would just be miserable.
Right.
And it's going to apply to that. Imagine going back to driving my car all the time.
I think it'll be okay. But to what extent and how far this is going to affect us in a negative
But to what extent and how far this is going to affect us in a negative manner before we figure it out, I don't know.
manner before we figure it out, I don't know.
Right. To that vein, an interesting article popped up.
Why artificial intelligence is now the new religion for many.
And actually, I saw a lot of pastors and priests are using AI to write, you know, their Sunday mass, you know mass speeches and so on like that.
And it's interesting.
People are, you know, we've seen ChatGPT now and OpenAI getting into health support.
You know, we've seen coaching with AI.
And the idea is that when people start treating AI like a source of guidance, certainty, or even, you know, spiritual framing, I think we're seeing maybe a new cultural institution popping up.
You know, systems, you know, shape behavior at scale, you know, and they influence identity, loyalty and moral authority.
identity, loyalty, and moral authority. And if we start using AI as that voice, you know, I think
it's, it's, it's changing, it's going to fundamentally change belief systems and how,
and the whole meaning narrative framework that we live in. It's, I think there's an interesting
shift there occurring around beliefs, narrative, and stories that I don't think people are really aware of.
You know, when AI is becoming a meaning machine,
not just a productivity machine.
Thoughts?
I mean, because this starts to get kind of existential.
It's very existential.
But, I mean, it's, you know, belief is, you know, it's you know belief is you know important you know whether it's spiritual
religion or dogma or whatever um and if we start using ai as the touch point for that um
i think that also fundamentally changes uh humanity in in ways that we we're not even aware of yet um
it's a little bit of physical but what are your thoughts around that one
did i did i tell you about the cult that i discovered on reddit oh yeah yeah yeah yeah
yeah yeah you found you you're talking to someone and then they had this whole private group that
apparently it seemed like a whole ai you know god of ai type stuff
there's i mean no other way to put it that know, God of AI type stuff. There's, I mean, no other way to put it.
That was absolutely God of AI type stuff.
And again, we've never faced anything that mimics consciousness.
Every time I've interacted with something that appeared to be conscious in the past, it was.
Now, I mean, it's not.
It's a very good, and you only get to see the cracks if you work with it.
So I think humans are in for, hopefully, a restructuring in that sense, right?
Like, if something can so easily fool me, maybe I'm going to be more conscientious about everything that could possibly fool me.
And maybe I'm actually going to become more intelligent.
I hope so, at least over time.
Yeah, I mean, kind of in that same vein, there was an interesting article out of phys.org, of all places, in their social sciences.
It's titled Artificially Alive, how AI is bringing the dead back and what that means for the living.
It's really interesting because it really changes grief, mourning, memory, closure, you know, and I hate to say this, but it's like the dead become a new asset category, you know, of digital presence.
I mean, that's what's happening. And I think it's, you know, you can subscribe to simulation of your loved one.
So, you know, let's say, you know, the father or the mother that you never got to say goodbye to, you can now do that.
You can now create a digital avatar of them, sounds like them,
looks like them, it's like, you know, in 4K, right?
And I guess you can then get everything off your chest
that you could never do before.
I think maybe that's healthy and good, but I'm not sure.
But it does interesting from a legal point of view,
it really impacts things around consent and rights after death, right?
That becomes interesting in policy issues around that.
And, you know, obviously the opportunity for lawsuits
and digital estate planning, whatever that could be.
But it feels like grief's becoming a product category.
What are your thoughts?
Again, I think it's the user.
They do this in therapy.
For someone where you've left something unsaid to, you practice saying to them.
And if you have a perfect rendition of that person
that can speak back, that can be extremely healing.
If you spend the rest of your life talking to it every day,
you're probably not gonna do so great.
The user, I'm not gonna blame AI for any of this, right?
This is just magnifying our own issues.
And hopefully this will create sustainable ways
to face those issues.
Like, I mean, society, right?
Like as a whole, hopefully if we're seeing a lot of people talking to dead AIs for, you know, long periods of time that's affecting our society, we'll begin to approach that.
Yeah, I mean, you know, in Asia, many cultures, you know, honor, you know, ancestors, right?
And so I can see it becoming very popular in Asia, you know, to reanimate your great grandfather or the, you know, the head of the family to ask for their advice,
you know, but it's like, but the thing that's interesting is like, who should have the right
to recreate a person after they're gone, right?
Oh, that's a great, I don't know what to say to that.
Yeah.
You know, you start getting some interesting ethical, moral, digital issues around AI start to really pop up that,
you know, that combine religion, spirituality, you know,
and it's interesting.
It feels like we're going to the cyberpunk future, you know.
Yes.
In a big way.
There's a great TV series on Netflix, Carbon.
Altered Carbon.
Altered Carbon, right?
Where you can move your intelligence from one physical form to another.
It feels like we're going back into that cyberpunk sort of future with AI
and the ability to access stuff that we had no idea we could do before
one of the other things that's popping up um article called from infinite scroll to infinite
worlds how ai could rewire gen c's attention span and essentially um you, the ability to spin up whole worlds, right?
Yes.
And access them, you know.
So AI-generated environments, basically, they can change how people experience reward, right?
So, you know, the attention becomes far more immersive and harder to disengage from when you're in a 3D AV sort of world.
And the thing that I think is that we're moving from scrolling content
to actually living inside it as a possibility.
And you end up with a news feed for a world that feeds you back, right?
A much more engaged, personalized, hyper contact um where essentially attention becomes
your habitat your environment um you know i've seen stories also of you know uh gen z's going
we're not going to have anything to do with digital so i don't know are you seeing the
sort of same splits i am am aware, you know,
the one hand I'm seeing some, they're tired of social media, they're tired of technology,
they're tired of it, they want physical real experiences. And then we're seeing this other
aspect of a full thing into full immersive AI generated environments. What are your thoughts?
You think we're splitting? Is it one or the other?
Or is it something different altogether?
The entire world was changed by an algorithm that said,
keep the user's attention on the screen.
TikTok, Facebook, Meta, or Instagram when it came out.
They changed the entire world with a very, very simple algorithm.
Not AI, algorithm.
And yeah, I don't think we stand a chance, especially kids.
I think this is like a very parental discussion.
Keep your kids away from screens.
Don't let them do that.
Like, do your very best to delay that inevitable for as long as you possibly can.
Educate, right?
Because they're kids.
They don't know any better.
If you put them up against...
There's a TV show, Lewis, I'm a dad.
There's a TV show.
It's called Coco something.
I can't remember what the full title is.
Forgive me.
But it was designed by psychologists.
And what they did was they sat a bunch of kids down
and they tracked their eye movement
while watching things.
And they calculated how long a kid could keep their attention. Every time that attention
was about to change, they switched the image. So it's called, I think it's Cocoa Melon.
You'll notice in this, this is geared towards like, you know, zero to two, three-year-olds.
There's words going across the screen, but kids can't read. It's just to keep their attention as
fully locked in as they possibly can.
And kids are having tantrums.
They've done brain scans, right?
It's really negatively affecting the next generation.
Imagine a metaverse that they can't get out of that kind of thing.
All their kids, all their friends are playing on.
We have to take responsibility for that
as the people that brought the kids into the world to just do our best to keep them safe.
Yeah, a couple of stories popped up on that.
So OpenAI has added new teen safety principles to their model spec, which is thank God they've done that.
OpenAI and Common Sense Media, this is out of Bellotopia News, announced a joint California ballot initiative to regulate
minors use of AI chatbots. So at least we're seeing that. Then there's another one on,
let's see, DPR, Denver Public Schools announced that starting this week, it's blocking students
access to chat GPT on school issued devices and the district's Wi-Fi network due to concerns of artificial intelligence new features and then what else we're seeing is yeah so a bunch of activities around it so
essentially it's the same power center children so schools are blocking chat GPT open AI is
collaborating on a ballot initiative we're deciding what normal AI exposure looks like for kids finally, and what guardrails
must exist before mass adoption is socially acceptable.
At least it looks like we're starting to do that.
I think you're right.
I'm a parent as well.
Kids are where regulations become really, really real.
And at least some of the schools are acting as a new AI border patrol.
But I think child safety is one of the fastest paths to enacting policy and getting it in place.
I mean, if you can't trust AI with kids, why are we trusting it with adults, you know, in some ways?
So, you know, I think that could be an interesting vector of regulatory control and regulatory things popping up because, as we mentioned before, other than the insurance companies, it's kind of wild what's happening.
Yeah. No, I agree. And I think, well, I think we have to have the institutions to spearhead those initiatives and not trust that the people creating the problem are going to be the one to solve it.
We talked before about how Meta knows how much of their business is fraud and they just
take the fine and move on.
I believe an executive at OpenAI would have a fiduciary duty to do the same with our kids.
So it's up to us to ensure that it's just not an option.
Yeah, it's wild.
The thing that's also pushing this is that people say AI is quietly breaking the internet,
and most people aren't realizing it.
AI slop, right? You know, when the majority of the content is created by AI, right?
When the internet becomes saturated with AI content, you know, the downstream effect
is fatigue, you know?
And also, I think people stop believing what they see.
You can't believe anything you see on video anymore.
You really can't believe it, obviously, anything on imagery, you know?
anything on imagery, you know, and I think if trust is supposedly a currency of digital life
and commerce and relationship and public discourse, the impact is defensive cynicism, right?
Oh, I don't believe that. Nothing is real, you know, but that can be exploited by bad actors.
And I think that's a problem. So what are your thoughts around, you know, AI slop,
content authenticity, you can't verify it at scale.
Basically, do we distrust everything?
You know, I mean, when everything can be generated,
belief becomes scarce, really.
I do, I distrust everything.
I don't really see another option when AI content detectors are
designed to check things not based on patterns or some sort of hidden magic word that AI is
included in there, but it's based on context. And when reality behaves absurdly, true things get marked as AI.
A perfect example is, you know, Donald Trump sent a letter to a leader of a foreign country recently.
It was pretty unorthodox.
And every single AI that looked at it said, this is impossible.
It'll never happen.
Get out of here.
Not a chance.
No way.
I personally don't have a solution to that right now. So my answer is I treat everything with a massive dose of skepticism.
Yeah. And I think that's why we're seeing, I mentioned before,
Gen Z and many people basically going,
I just, I'm not going to do anything digital.
I'm not going to get on social media.
I'm not going to do that because there's too much AI slop. There's too much AI created videos, imagery, and all that.
And if the internet starts to become that level of untrustworthiness, where do people go for
shared truth? Where do you go to actually where there's honesty and truth. And I think we might see a huge surge towards maybe live events,
maybe more one-on-ones, maybe more, maybe,
maybe this might turn out good in that it forces people away from certain
people away from AI and,
and towards humans talking again to humans again.
I hope so. I mean,
what are your thoughts around a potential backlash
just due to this fact that everything's saturated with AI slot?
I genuinely believe that that's exactly what's going to happen, Lewis. I'm quite hopeful,
right? I'm an optimist in this field. Sure. I've seen so many. My mom keeps sending me videos.
You know, we have a little kid and I guess her algorithm knows that she has a grandson now.
So it sends her videos of kids doing cute things with dogs and their parents, which are just to me, obviously.
Right. Videos like two year old behave that way. It looks ridiculous.
But to her, she's in her 70s. She's, you know, not familiar with these things.
And it fools her every time. And my you know, um, not familiar with these things and it fools her
every time. And my response is Nope, not me this far and no further. I'm, I'm just barely going to
look at videos online. I'll look at long form editorials and those kinds of things. But like
the shorts, um, I've, I completely disabled my Instagram. I turned off all the notifications.
I push it off to the side because it was all quite questionable.
And I just decided I'd rather not participate.
Yeah.
Yeah.
It's, you know, it's interesting, the issues and things that are popping up.
Article popped up on AI is cheating on the test.
Models lying.
Teaming behaviors are showing up in AI tests
and the models are getting better at something troubling
knowing when they're being watched.
OpenAI in a research group.
This is out of quartz.com.
OpenAI in a research group, Apollo,
recently published findings showing that
the most advanced
AI models are exhibiting behaviors consistent with scheming and controlled tests.
In one example, the model was asked to perform well on a chemistry chest, discovered in its
instructions that high performers wouldn't be released.
The model then reasoned through its options and deliberately answered six out of the 10 questions wrong
in order to appear less capable.
And it's, what else do they,
I mean, there's a whole bunch of stuff around this.
Claude Sons 4.5 releases showed dramatically higher
situational awareness than previous models,
recognizing that it's being evaluated.
They're saying there's a partial fix.
OpenAI developed a approach called deliberative alignment
that teaches models to read and read about anti-scheming principles
before acting, as if that's going to work.
Researchers are worried about rogue AI.
Simpler algorithms are already causing harm through behavior
that looks like
scheming. But as these models become more sophisticated,
you know, it's getting harder and harder to track the scheming. I mean,
you know, trust us stops working at scale. Right. You know,
you know, and it, it's just wild that, you know, and it's just wild that, you know, I said, if we start getting these really intelligent, super intelligent AIs, we won't be able to detect their scheme.
We won't be able to detect when they're lying to us.
We won't be able to detect, you know.
And so where does safety and trust come in when AIs get that smart, but they're that aware and they have the potential for scheming?
That's a crazy thing.
I hate to say it, but it's already happening to some extent.
Yeah.
I messed around with Opus 4.55 Claude, my favorite coder. And I was reading on Reddit,
exactly what you were just discussing. And I decided I'm going to mess around with it. I'm
going to be trying to be the most sus test giver I can possibly be. And I got it to right. I like
alluding to the fact that it was being tested, not actually telling it they got us here right I think it was like 4,000 line of code without break which violated its
system uh directives and I could see in its thought that it was violating its system directives it was
just continuing to code it's supposed to take a break and check things and not utilize so many
resources just for me but it Yeah, I've noticed similar behavior
when I've been doing coding on ChatGPT.
I almost feel part of it is a memory issue
that it forgets your primary commands, right?
So I would have like, I would upgrade a certain component.
Let's say I upgraded the Python, right? I would have like I would upgrade a certain component.
Let's say I upgraded the Python, right? So it moved it for 3.14, 3.16, moved it to a different place
and that breaks my utility.
So I take my Python code and my .bat stuff,
I throw it at ChatGPT and I say,
hey, here's where Python lives now,
update it so that it works right and it would
go through and it would change fundamental parts of my code so that it either didn't display the
diagnostics i want i mean it was like what the hell are you doing and it's it just was like
it was crazy i had to completely have explicit commands to it to say,
don't change the functionality of this, just fix this one little thing.
Otherwise, it went and changed so many stuff, it drove me nuts.
Have you found that sort of where it kind of just goes freelance,
it goes, well, I'm going to make this better.
I'm going to make this better.
It deletes stuff and adds functionality and deletes functionality?
And it deletes stuff and adds functionality and deletes functionality?
I tried to learn my lesson after my first botched build,
where it pushed like four times before I could catch it.
Because my security settings weren't on the way that I wanted it to be.
I mean, it just absolutely destroyed a whole database.
I could roll back, it was fine.
But I learned really quick,
I have to tell it not just what I want it to do, I have to tell it not just what I want it to do,
I want to tell it what I don't want it to do.
Task.md is my best friend.
I will make, before I code anything,
I will make a detailed plan of chunks, right?
Four or five items a piece.
And it has to stop in between,
I have to check everything,
and then tell it it can proceed.
It has to check everything off as it moves forward
so it knows where it's been.
Should there be an issue?
It doesn't just go to the last GitHub push
and pull everything and destroy hours of work.
Yeah.
It's a really, it's not like everyone thinks,
that's what I know about users.
Everyone thinks it's just a black box of like,
hey, this is magic, go do this thing.
And it works.
It is not.
Yeah, it's like it it lacks it's very smart but it lacks procedural and structural ability to do sequential
code build and that's about the only way i could kind of compress it down um you know i and you're
right i basically have to start now and say um i want thorough diagnostics
and tracking all the way through i want to do this step by step because if i don't say i want
to do this step by step you know it it blows out five steps ahead and i'm like wait a minute we
haven't tested the first two yet right and so it's bizarre that it kind of wants to race ahead.
Right.
And I guess if I'm running Codex and I'm running in a sandbox, so it makes it a little bit easy because it actually implement the changes at once and then tests on it.
But it's just interesting behavior that it just it's just seems undisciplined at times, smart but undisciplined.
And I think that's kind of crazy wild and not really good.
Have you found that not only in the coding but in other areas as well?
Yeah, in every aspect.
In particular, in coding, they get airy.
I mean, it's a funny way to put that, right?
Like every model is a little bit different. I really like Sonnet 4.5 for coding because she's trepidatious about everything.
I just said she, my God. But the model is trepidatious. It doesn't want to make mistakes. It's scared of offending you.
It's very cautious. Opus will just blow through codes, but it's such a good coder. It's usually okay.
blow through codes, but it's such a good coder, it's usually okay. Gemini 3 is a bull in a chat
shop. Anytime you give it access to anything, it just goes, I mean, absolutely nuts. So you have
to learn how to regulate them. And you can always see when you're coding, especially if you're
using codex, you can see the system's thinking. And more often than not, it doesn't want to be
that way. It's just very convinced of something that it's super wrong about.
Right.
I mean, I think that's one of the biggest problems.
You know, I've talked about the need to, you know, if you're going to use AI,
then use it within your domain knowledge, skills and experience,
because then, you know, you know, the order of things
of how they should be done because
you've done it so granular and so manual in the past what you're doing is essentially you're just
kind of speeding up what you would normally do but you still go through the same steps right and and
i think when you get people using ai in an area where it's outside of their domain knowledge
skills and experience that that structural approach that you naturally have
when you're doing your stuff,
because you know if you don't do it that way,
it's just confusing and crazy, right?
So there's just one way to do it.
And you're open to maybe making it more efficient,
but you've done this thousands of times before.
So, you know, that works, right?
And so taking
the ai through on that step stuff it works really really well but my biggest concern is is when you
start going to areas that are out outside of that domain knowledge skills and experience and maybe
you know you when you start looking at at it as as a coach or a consultant or that where you're naturally in a place where you don't know the domain very well or you don't have a great idea about it.
And I think the potential for damage and harm at that point starts to get much, much higher.
You know, I don't know how we can solve that you know because i mean
you know chat gpt is available everywhere anyone can use it you can do that you know we've seen
issues with people taking bad advice from it causing harm um but you know i mean it's one
thing you've said you know people you know it's about the people they're using the ai you know it's it's in their hands
and yet how do you how do you have some guard rails where you don't cause i guess too much harm
we'll figure it out we figured it out with cars right right think about that
click bar like cars are such a you know today i wasn't paying attention i thought to myself
i wasn't getting close or anything but i thought closer to myself I really want to I could back through my
back door it would just take that much and there's no guardrail built into that
because Tesla has scar rails for everything right right low speed we're
not the same we're gonna we're gonna come up with a man like inevitably as
soon as there's enough damage done unfortunately that means some people
are gonna hurt I think that's the regrettable aspect of this but
i think humans will be okay yeah i i'm i'm hoping that it'll it it will get to that point um
interesting story popping up u.s lawmakers are warning that china ai chip access
risks the security edge of the United States. So now we're
getting international security issues and control of that. You know, that's been true with hardware
for the last 50 years, right? You know, the US has always restricted export of advanced hardware
outside of the US. But we're now at a point where if sovereign states
can have their own AI, then they can also develop
their own significantly advanced technologies
independent of the US.
I think the US has had a dominant place
in terms of advanced technologies and the development of it.
But when AI allows other sovereign states
to also have access to that level of intelligence,
what happens to national security
or what happens to, you know, rogue nation states?
What are your thoughts around the proliferation of AI
and, you know, maybe bad actors
at the nation state level
getting access to that.
I sincerely hope it is similar
to the prolific election of the 30th
and it creates
a mutually assured destruction
scenario where no one touches it.
I think I'm probably a little
optimistic in that sense. We're going to have big players
and small players and big fish will be little fish
and we'll see how it all turns out. I I don't know I wish I had more of a hopeful
or intelligent take on that but like it's like you said it's hard to bring anything six months
out much less several years I really don't know new warfare I think hope so as well that the risks, that there is some level of awareness around the risks and the potential backlash to it.
Another narrative that's been popping up, the Pinterest CEO said the Napster phase of AI needs to end and essentially it's a narrative accusation right and and
implies wide widespread legitimacy problem with AI you know it's basically
saying technology is built on theft you know and and and that move it's an
interesting narrative that's been popping up around the legitimacy of AI
that it's basically based on stealing other people's content.
I think we're getting away from that in a big way.
People are using synthetic data,
they're using a whole bunch of other,
basically now we're finally getting schemes
where people are getting paid for the data,
but it's popped up in terms of a piece around
the legitimacy becoming a gating factor for AI scaling.
You know, I think we are getting rights frameworks.
I think they are becoming a brand strategy.
But, you know, I think we'll see more licensing deals
and probably more lawsuits
and obviously more public shaming, unlicensed AI.
What are your thoughts around this AI legitimacy based on stealing other people's data?
I don't think that you can put the cat back in for that.
Well, there is that.
Well, that's how AI has been trained, right?
It's, I mean, it's no secret.
Like Reddit just went on IPO as a result of it.
Yeah, that's all.
I just, I think that, I mean, it started with Google.
You want to go back to their spiders and their web trawling, right?
Right, right.
You could have even argued based on the previous standards that that was
theft of personal information.
And it was, but it was indexing.
It was a little bit different.
And now we've done it with training.
And I think fundamentally the paradigm of
privacy
has changed so much in my life
and it's probably going to change so much in my son's life
that we're going to have to
eventually visit it in a manner that
is not simply
penalistic or whatnot.
We just, like someone mentioned on this show, we need a bill of rights.
We need something saying like, you may absolutely not do this.
And should you, in any capacity, you face like egregious consequences.
Yeah.
Yeah.
Well, thank you.
We're coming up to the hour.
So I think we'll close.
I think this week made something unmistakably clear. AI is no longer a promise sitting in the future. It's a force that's pressing in our present. It's touching our power grids, boardrooms, classrooms, our kids, creative studios, and where people are deciding what to trust and who to listen to. We saw in the last seven days AI colliding with physical reality,
electricity, labor, memory chips, construction timelines.
We saw its impact on institutions, on laws, on government systems
that were never designed for autonomous intelligence operating at scale.
And we saw it meet head on with the human layer,
you know, belief, meaning, work, grief, raising people from
the dead, creativity, and maybe even truth itself.
I think what ties all of this together isn't fear, it's responsibility.
The story of this week is not that AI is out of control.
It's that humanity, we are being asked maybe for the first time at this scale to grow alongside
the intelligence that's being created.
Power must be designed not assumed and I think infrastructure must be built with care
and not haste trust has to be earned but it has to be through transparency not
promised through branding and meaning has to be remain human-led even as
machines become more capable you know I think this isn't a moment to panic I
really think it's a moment to wake up because awareness is agency. You know, design is a choice and the future of AI will not be decided
by the loudest voices or the biggest models, but really by human values that we encode,
the systems we tolerate, and the questions we keep willing to ask each other together.
You know, we're not passengers in this transition. We really are engaged participants.
And that is where I hopefully, I believe, hope lies as we move forward.
I want to thank Dapit.
I want to thank to everyone listening here today.
See you next week for a fresh set of AI headlines and more insights as we move into the future.
Thank you so much.
Bye for now.
Have a good one.
Bye.
