Tech leaders unveiled the American AI Action Plan on Wednesday and President Trump signed 3 executive orders with big handouts to AI companies. Madeline returns fresh off of a trip to Washington DC and gives Tom and Eric the lowdown on tech's victory lap at the nation's capital. Plus, even the altruists at Anthropic feel the need to raise Middle East money.Timecodes:01:45 - The All-In podcast might as well be state-run media05:16 - The vibes at the All-In Hill and Valley event08:50 - The coming AI abundance12:29 - Woke AI and the ministry of truth15:30 - America's strategic advantage is President Trump22:36 - The AI copyright kerfuffle32:36 - Dario faces the harsh reality of capitalism
00:00:00
Hello. Hello, everybody.
00:00:01
It's Tom Doughton here bringing you their latest episode of the
00:00:04
Newcomer podcast, joined by Eric, Newcomer of Newcomer and
00:00:08
Madeline Renbarger of Newcomer. Welcome everybody.
00:00:11
How you guys doing? You love that.
00:00:13
I'll never stop. I asked Tom if you wanted to
00:00:15
change the podcast name. He said no.
00:00:18
It's a brand. It's a big giant end.
00:00:20
Well, I'm just a contractor, so I'm actually a free radical here
00:00:23
You guys are deep in deep in the swamp of newcomer.
00:00:25
Speaking of the swamp, Madeline, you just returned from the
00:00:29
swamp. I was talking to you yesterday.
00:00:31
You were at a defense tech conference, Hill and Valley.
00:00:34
Why is it called Hill and Valley?
00:00:36
So I actually let me clarify, it was Co hosted with the Hill and
00:00:40
Valley forum, but it was actually called winning the AI
00:00:43
race, which was a new conference also Co hosted with the all in
00:00:46
podcast. So I want to clarify for the
00:00:47
brand's sake, a different event, but it's called Hill and Valley
00:00:51
because it's, you know, the merger of, you know, DC, the
00:00:54
Hill and the Silicon Valley coming together.
00:00:58
Silicon Valley. See, I was thinking, no, I'm not
00:01:01
smart in that way. And I actually I was liking my
00:01:04
brain because you know, there are some like Chinese based
00:01:06
private equity firms called like Hill House Capital and stuff and
00:01:10
you never know what that means. I thought this was some weird
00:01:12
permutation of that, but no. This podcast is supported by
00:01:16
Google. Hi folks, Paige Bailey here from
00:01:18
the Google DeepMind Devrel team. For our developers out there, we
00:01:21
know there's a constant trade off between model intelligence,
00:01:24
speed and cost. Gemini 2.5 Flash aims right at
00:01:27
that challenge. It's got the speed you expect
00:01:29
from Flash, but with upgraded reasoning power.
00:01:32
And crucially, we've added controls like setting thinking
00:01:35
budgets so you can decide how much reasoning to apply,
00:01:37
optimizing for latency and costs.
00:01:39
So try out Gemini 2.5 Flash at aistudio.google.com and let us
00:01:43
know what you build. Tell me, tell me so and you just
00:01:46
found out at some point as you were there like, ah shit, I'm at
00:01:49
an all in conference. Yeah, so it was advertised as
00:01:52
being Co hosted by All In. But what I did not realize
00:01:55
getting into it is that the on stage programming was a 5 hour
00:01:58
live all In podcast. Where the All In.
00:02:01
Guys, I didn't even realize that.
00:02:03
Yes, they all in guys were on stage with all of these tech
00:02:06
leaders and Trump administration officials just kind of running
00:02:09
through Trump's new AI policy and reacting to it and chumming
00:02:13
it up and hanging out on a couch.
00:02:15
That was the general vibe. There also was the classic
00:02:18
networking section outside where you could talk with founders and
00:02:21
investors that attended. My God.
00:02:23
So we're gonna have a whole future media segment of this
00:02:26
show, I guess. Man, I didn't realize that piece
00:02:29
of it. Or I guess we have to make like
00:02:31
Cerebral Valley a live podcast, like a 5 hour podcast now.
00:02:35
Like if that's the bar, we. Just want like filibustering but
00:02:38
anyway. Well, it's kind of funny to me
00:02:40
that I mean, you know, not to start off with the all in guys
00:02:44
because really the less about them, the better we say.
00:02:46
But like, you know, they've played a huge part in, you know,
00:02:49
the the Trump, the tech MAGA movement.
00:02:52
And obviously David Sachs is part of the administration now.
00:02:55
But this seems to be like the fullest, like the most fullest
00:02:58
expression of like their Trump ties, right?
00:03:02
This is like the full the first time that they've done like a
00:03:04
conference around the fact that they are like deeply embedded in
00:03:07
the administration. I mean, David Sachs is working
00:03:10
for them at this point. Like they've had conferences.
00:03:13
I don't know. If anything, I feel like some of
00:03:15
the like the sex appeal and freshness of it all is going
00:03:18
away. It's like they've done
00:03:20
conferences. Trump's there while the Epstein
00:03:24
files are dropping. It's like, it feels a little bit
00:03:27
like, I don't know, state TV to me where they're like on the
00:03:31
team setting it up. Did did they bring?
00:03:34
I assume Epstein did not come up at all on stage, right?
00:03:37
You know, Eric Epstein did not come up on stage.
00:03:40
And but their media figures on some level, like what media,
00:03:44
even conservative media figures, can't stop talking about Epstein
00:03:47
because it's such a good story. And so I think the fact that All
00:03:50
In would have a conference and not talk about like the biggest
00:03:54
elephant in the room. It we learned that day that the
00:03:57
attorney general had told Trump that he was in the Epstein files
00:04:00
and they don't bring it up. It's just that is the smoking
00:04:03
gun that all in is in no sense media.
00:04:06
But they're not not just not media.
00:04:08
They are not working on the behalf of their audience, right.
00:04:11
If you're working on the behalf of your audience, forget.
00:04:14
I don't even want to say I'm a journalist anymore.
00:04:15
You know, my whole thing has been like, I'm a creator that
00:04:18
wants to exceed expectations, but at least I know that I am
00:04:21
working for my audience and the readers.
00:04:24
And if they had an obvious question, I had someone up on
00:04:27
stage, I didn't answer it. They should make fun of us, you
00:04:31
know, and all in is is working for their subjects and not for
00:04:36
their listeners, whether their listeners want to acknowledge
00:04:39
that or not. And I think this event was sort
00:04:41
of the capstone of that sort of capture.
00:04:46
Well, I feel like they're listeners and their audience
00:04:49
and, you know, I feel like it's kind of merging into one, right?
00:04:53
Like if this is the culmination of, you know, all in post having
00:04:56
Trump on the podcast, they talked a lot about how, you
00:04:58
know, they saw the light after he was such a great guest on the
00:05:01
show to then, you know, be parts of the administration.
00:05:04
Well, Trump. Trump literally thanked even
00:05:07
Jason. Even as we know, Jason
00:05:10
Calacanis, I say even thank you, Jason.
00:05:16
All right, in in classic fashion, we couldn't hold off on
00:05:19
takes for long. So I've already given you a rant
00:05:21
without even laying the groundwork.
00:05:22
But Madeline, can you give it give us sort of a little bit of
00:05:25
the overview of the day. Yes.
00:05:27
So spearheaded by the all in guys, Trump was obviously the
00:05:30
draw. You did what the quick trip to
00:05:34
DC from New York, train there train back same day.
00:05:36
What was what was on the ground who talked?
00:05:39
Just give us sort of a couple bull points on what it was like.
00:05:42
Sure. It was definitely, I think
00:05:44
compared to past Hill and Valley forums, which I've covered much
00:05:47
more emphasis on government officials.
00:05:49
There were, you know, Chris Power from Hadrian who just
00:05:52
raised a big amount of money, gave a presentation on stage
00:05:55
about how they're going to, you know, build up AI factories.
00:05:57
And Jensen of course, was on stage with the all in guys
00:06:00
talking, you know, thanking President Trump for accelerating
00:06:02
AI and talking about all of the great jobs that AI build up is
00:06:06
going to create. However, most of the people on
00:06:08
stage were predominantly administration officials.
00:06:11
So it felt a lot more like here we are, it's DC, we're coming on
00:06:16
tech friendly media to talk about all the great stuff we are
00:06:18
doing for you, the tech audience.
00:06:20
And this was literally like an announcement, right?
00:06:22
I mean, they literally published winning the race America's AI
00:06:26
Action Plan, like right around this.
00:06:28
It was. The same day, about an hour
00:06:30
before the event kicked off. And Davis X is the AI czar
00:06:34
bringing in his in house media and basically like.
00:06:37
But that's just smart event planning.
00:06:39
That's just called synergistic thinking.
00:06:40
Democrats need to get this shit together.
00:06:42
Have friendly softball questions from in house media you know.
00:06:46
Oh, I just meant like having an announcement, you know, you want
00:06:48
news to break on the day of your conference, you get people
00:06:50
talking. I mean, good, good on them.
00:06:53
That's actually sounds like a very traditional media playbook.
00:06:55
And it seems to me like they're really in the pocket of
00:06:56
traditional media now. Yeah, what I heard Madeline was
00:07:01
saying that all the reporters were passed off to the balcony.
00:07:04
You couldn't even hear the toss A.
00:07:05
Member As a member of the New Media Madeline, how how are you
00:07:08
treated? As a member of the new media, I
00:07:11
felt like I should have been treated a little better since
00:07:13
we're, you know, not the lamestream media over here at
00:07:15
Newcomer I, the press area was a little bit separated off from
00:07:19
the main auditorium, up on a second level.
00:07:21
They didn't really want us coming down to the main level
00:07:23
to, you know, mingle with people unless we were in the lobby, you
00:07:26
know, grabbing coffee and stuff. So I do feel like the the media
00:07:30
was kind of in attendance, but not really participating in the
00:07:33
the same way as past events that I've attended.
00:07:36
You should have you should have like a flat, you know, like the
00:07:38
the press, you know, flak jackets that they get like when
00:07:40
they're embedded people. I feel like you could get a
00:07:42
separate one that says sub stack on it and they just treat you a
00:07:45
little different in these events.
00:07:46
I think I should have gotten special treatment because we're
00:07:48
a sub stack newsroom. I think we are the independent
00:07:51
media that they want to have covering this.
00:07:53
You know, we're, we're breaking the molds.
00:07:55
Sub stack needs to have his own advocacy org, you know, just
00:07:59
like there's like the White House Correspondents
00:08:01
Association. The sub stack should have sort
00:08:03
of a centralized like you can't treat our people like that.
00:08:06
But OK, I know it's Trump and like it's mostly optics and says
00:08:10
a bunch of insane shit. He talked about like maybe we
00:08:13
should break up NVIDIA. And then somebody explained why
00:08:16
that would be a terrible idea given Nvidia's like the heart of
00:08:19
everything we're doing. But so there was lots of insane
00:08:21
Trump shit and we can talk a little bit about like, you know,
00:08:24
his demeanor, but the substance, please, just what what?
00:08:29
Just just the substance, just quickly, what what actually
00:08:32
happened? Like what they wield all the
00:08:34
levers of power, Like all of them.
00:08:37
Like what? What are they doing?
00:08:39
Yeah. I mean, they had on stage the
00:08:41
Vice President and the President of the United States.
00:08:44
What are they doing? Like what is?
00:08:46
What are they? Doing not.
00:08:48
Saying like what is going to happen?
00:08:50
The key thing that is going to happen per the executive orders
00:08:54
is that environmental regulation is going to be pulled back so
00:08:57
that this data center build up will be easier to do.
00:09:01
Fuck the environment. Build AI, get energy ready.
00:09:03
Let's go. Yeah.
00:09:04
It was very build, baby build, you know, instead of drill, baby
00:09:08
drill, there was build, baby build.
00:09:09
That was the vibe inside the room.
00:09:11
Clear regulation, clear cut regulation.
00:09:13
So AI can like dominate and we're going to have this energy
00:09:16
crisis. So we need to light every like
00:09:19
dead dinosaur that we got left on the planet.
00:09:22
Sure, the abundance guys are shedding a tear of joy in the
00:09:25
midst of all of that so we can find.
00:09:26
Some, yeah, it's pretty abundant.
00:09:28
We're going to have AI abundance.
00:09:30
Step 2, the second executive order really touched mostly on
00:09:34
around exporting AI to make sure that American AI can be
00:09:39
exported. So that's, you know, the plug to
00:09:41
Jensen. We're not going to totally clamp
00:09:44
you down with export controls. After all that China fear
00:09:47
mongering, it's like actually, after creating tariffs and
00:09:52
saying isolationism is the path to American success on AI,
00:09:56
somehow they've conned Trump into saying actually America's
00:10:00
AI succeeds if we can sell it everywhere.
00:10:02
So it's embedded even in China. We want NVIDIA chips in China
00:10:05
because otherwise it's going to be Huawei.
00:10:08
We are globalists and internationalists on AI and then
00:10:12
randomly on like manufacturing cars and any industry that
00:10:16
doesn't have a great lobbying firm where isolation is.
00:10:18
It makes zero sense. Oh, that's totally the policy.
00:10:22
Well, I was going to say too, I think what was really just, you
00:10:26
know, the tell of all of this was that during Trump's speech,
00:10:29
you know, he bragged about all of the tariffs, making jokes
00:10:31
like there's too many countries. I have to do deals with all
00:10:34
these countries and, you know, touted his recent Japan deal.
00:10:37
But then of course, with AI, it's like, cut it loose, let it
00:10:40
go because that's, you know, American soft power ruling the
00:10:43
world or else China AI will take over.
00:10:44
So I feel like there were the inconsistencies in the policy
00:10:47
really kind of popped off throughout the day.
00:10:50
Have you just paid a little bit of attention?
00:10:52
I obviously believe in some environmental regulation, but I
00:10:54
think clear cutting regulation to build AI is good.
00:10:59
I think for the if you buy into the like the Super growth
00:11:04
mindset that we need to do this in the China race, it certainly
00:11:07
is good. I do think that the extent that
00:11:09
they want to cut it is definitely, you know, no gloves
00:11:13
at all, just build drill. But we're going to be helping
00:11:16
China now. We're selling them NVIDIA chips.
00:11:18
Like, I don't think it's about like the war with China.
00:11:21
Like I don't buy it at all. But we're also beating them.
00:11:24
Sure. We're beating them.
00:11:26
We're giving them the tools to undercut us and steal.
00:11:30
Yeah, I want to cooperate with China, so I I have no problem
00:11:32
with it. Godspeed.
00:11:33
But the idea that it's going to help us beat China by giving
00:11:36
them the chips that they need to deliver the stuff doesn't make
00:11:40
sense to me. But also the export ban wasn't
00:11:42
working like the the the whole thing end up being they were
00:11:46
getting in there through like Indonesia and like various
00:11:48
Southeast Asian countries like you saw with DeepSeek that they
00:11:51
were plenty capable of building like competent high quality
00:11:55
models largely on chips that had been smuggled into the country.
00:11:57
So. All it was really doing was
00:11:59
hurting American chip manufacturers, which is why
00:12:01
Jensen was kind of pushing them to, you know, hey, don't, don't
00:12:04
lock this in this way. Can I ask quickly because I, I,
00:12:08
I saw the tweet, I didn't dig more into it.
00:12:11
Why did he want to break up NVIDIA?
00:12:12
What was the argument there? Why were they scary to him?
00:12:15
I might be misremembering, but it felt like it was like an
00:12:18
NVIDIA. Who are those guys?
00:12:20
What are they doing? No, they're American.
00:12:22
Asian guy that runs it, that's no good.
00:12:25
Break that thing up, but it's like.
00:12:27
You can't tell me that there wasn't at least an initial
00:12:29
thought by Trump that like they're a Chinese company.
00:12:32
Right. No, literally.
00:12:34
On stage. He more or less implied that in
00:12:36
his little NVIDIA riff he was doing in the speech.
00:12:39
Like, so there there's this AI action plan, but then there were
00:12:43
also executive orders about AI, the most disturbing of which to
00:12:47
me was this order that says if you're a government contractor,
00:12:52
you have to have unbiased AI. And they specifically talk about
00:12:57
DI and transgenderism. And it's just like, yeah, I
00:13:02
mean, the idea of unbiased is incoherent.
00:13:05
And then in the same executive order, you're basically saying
00:13:08
what you can't talk about. So it's like inserting bias.
00:13:12
I, I don't know, I find that deeply troubling.
00:13:14
The last thing I'll say on this, I, you know, one of my more
00:13:17
viral tweets, I said, you know, no atheists in a foxhole, no
00:13:21
libertarians in a bank run when David Sacks was like begging for
00:13:25
SVB to, to get a bailout. And now it's like, you know,
00:13:28
there are no, there are no free marketers in government.
00:13:31
It's like as soon as David Sachs touches the government, they're
00:13:34
putting out executive orders saying let's let's regulate AI
00:13:38
so that you can't explain transgenderism to people.
00:13:42
Well, it's always just the point of, you know, the free speech
00:13:45
for me and not for thee. We saw this with the social
00:13:47
media wars, right? Like this felt like kind of
00:13:49
social media Part 2 where Zuckerberg caved and, you know,
00:13:52
got rid of community notes and Meta and all of this stuff to,
00:13:55
you know, get these get close to Trump and buddy buddy with the
00:13:57
administration. And so by just declaring that
00:14:00
for AI companies to work with the government in any way,
00:14:03
especially with all these DoD contracts on the line, they have
00:14:05
to, you know, remove bias just like that just means that they
00:14:09
have to censor certain topics because the training data is
00:14:11
already in there. Like what do they want?
00:14:12
And of course, which AI company has seemed the most biased so
00:14:16
far? You're talking about the one
00:14:17
that has written into. It's like prompts that like.
00:14:20
What would Elon say on this topic?
00:14:22
Right, the one that literally talks in first person sometimes
00:14:25
as the owner and searches like what's the guy who owns me think
00:14:29
before I answer this. Well, I also love the idea of
00:14:33
like, you know, a contractor having to explain to whatever
00:14:36
the DoD or any other organization that's going to
00:14:38
sign a huge, you know, trying to get like a huge RFP for the
00:14:42
different language models. Like Sir, we have found the most
00:14:45
unbiased of all of the chat bots that are out there.
00:14:48
Side point, it is Mecca Hitler. So you you will have to work
00:14:52
through that. But yes, as we ran through it,
00:14:55
the most unbiased 1 happens to also give specific instructions
00:14:58
on how to attack Will Stancil in the night.
00:15:00
But it is the least woke. So that is the one we're going
00:15:03
to have to choose based on the EO.
00:15:05
Yeah, it is just the hypocrisy of, you know, bias just meaning
00:15:10
different from Trump's positions on issues like that is if we're
00:15:13
just classically going back to that, it's just the greatest
00:15:16
hits of that. Again, we're just putting that
00:15:17
in executive order. And I just, I'm curious to see
00:15:21
how companies respond to it because I mean, working with the
00:15:24
federal government has proven to be very lucrative for a lot of
00:15:27
these companies so far. So I don't know what's gonna
00:15:30
happen. Well, we, we, we also saw like
00:15:32
in Trump one like vendettas being used as a way to hand out
00:15:36
contracts or stop other companies from getting contracts
00:15:38
was like fairly standard procedure.
00:15:40
Like Amazon AWS lost their shit after the Pentagon gave a
00:15:44
contract to Azure and to Google over AWS because of, you know,
00:15:48
Trump's anger at Bezos and, and Amazon and the Washington Post.
00:15:53
And so, yeah, no, I'm just saying like there's a real
00:15:55
reason for these people to like, play up at, you know, Curry
00:15:58
favor and like be as sycophantic as possible to Trump because.
00:16:01
Oh my God, yeah. Didn't Jackson literally say
00:16:03
America's going to win an AI because of Trump or something?
00:16:06
I America's unique advantage that no country possibly have is
00:16:12
President Trump. There there are a lot of
00:16:16
compromises I wouldn't want to make as a public company CEO.
00:16:19
Like, I don't know, but like the Oh yeah, Trump, you're a genius.
00:16:23
Like everything through you. It's like it's such a low
00:16:26
hanging way to Curry favor that I can see why it's so tempting.
00:16:32
And Jensen smartly, you know, got his export controls
00:16:35
overturned, so it's a good play. His approval ratings and Tom,
00:16:37
you're making this point before the show.
00:16:39
They are bad, like, somehow, like, until we have the
00:16:42
midterms, I feel like the culture, you know, if this were
00:16:44
like a normal president with approval ratings this bad,
00:16:47
people would, like, start to not taking him seriously.
00:16:50
But I think it's because the Republicans would get primaried
00:16:53
anyway that they have to be so locked up.
00:16:55
I don't know. Yeah.
00:16:57
Tom, what's your observation on the.
00:16:58
I think. We've talked about this a couple
00:17:01
times on the show, like, you know, the bald faced sycophancy
00:17:04
of the tech, you know MAGA, right?
00:17:06
Whatever you want to call it toward Trump at one point seemed
00:17:09
fairly prescient, right? If you were, I don't want to
00:17:12
talk about Sean McGuire again, But if you were one of those
00:17:13
people that were backing Trump during the 2024 election, you
00:17:18
looked really smart, like you looked way ahead.
00:17:20
Well, I would say to this point, Tom, the all the Hill and Valley
00:17:23
Forum leaders who, you know, put on this summit in the 1st place
00:17:26
with Jacob Helberg, you know, kind of spearheading this with
00:17:29
Delia and and Christian Garrett, they have, you know, hold the
00:17:32
anchors of powers in some way. And you know Helberg is going to
00:17:35
be appointed in the administration.
00:17:37
Republicans, if you're willing to be the one smart person
00:17:40
around a Republican, you could like, fucking ascend to the top
00:17:43
immediately. It's like it's a vacuum.
00:17:44
It's the easiest hack. It's like so many fucking
00:17:47
lawyers do that. But that's under the assumption
00:17:50
that being smart helps you as a Republican, which I don't really
00:17:52
think it does. I mean, but they always want.
00:17:54
I don't think to dominate. I mean, you see.
00:17:56
Yeah, you can. I think they can tell you the
00:18:00
story that being quote UN quote smart doesn't necessarily make
00:18:02
you the like the popular guy or the most powerful person in the
00:18:05
party. But you know, like Sam Altman is
00:18:07
an interesting case with that because he lifelong Democrat,
00:18:11
you know, is as public about hating Trump as almost anyone in
00:18:14
tech after he gets elected. I mean, he makes as a source
00:18:16
told me like not too long ago, he made a determination like
00:18:19
midway through the campaign, 24 Trump's going to win.
00:18:22
I need to start getting closer to him and he starts doing that
00:18:25
and then he streets out, you know, Oh, I misunderstood you,
00:18:27
Trump. Everything I thought, you know,
00:18:29
was, you know, intermediated by a biased media and all this
00:18:33
bullshit that everyone has to say in order to make themselves
00:18:35
feel, you know, close to Trump. So but that all seemed very
00:18:38
smart. And now, like, I was talking
00:18:39
about 37% approval ratings as low as it's ever been for him.
00:18:43
It's not as low as it got during Trump won, but like, he is not a
00:18:46
popular president. And and, and so when you see,
00:18:48
and it's not, I don't know if it's going to get better or not,
00:18:50
but it's unlikely to turn around in a major way.
00:18:53
And so when you see the same rhetoric from the tech crowd
00:18:57
supporting him, like Jensen saying, you know, we couldn't
00:18:59
have done without you, Trump, the big advantage that we have
00:19:01
is that you, we have Trump and no one else does.
00:19:04
It's so blatant as to what's going on here.
00:19:06
You're not tapping into a zeitgeist.
00:19:08
You're just trying to get the best possible deal for your
00:19:09
company. You're trying to line your
00:19:11
pockets. And that's fine.
00:19:14
That's business. It's it's gone like that.
00:19:15
Forever they want to get rich, they want the companies to do
00:19:17
well and that that's their job. But I, I do think they also want
00:19:20
to see, they think AI is like a bigger force than honestly, like
00:19:24
the, the US government and like the US government has a lot of
00:19:27
power. But it's like we need to suck up
00:19:29
the Trump and he's going to like clear cut the government and say
00:19:33
we can do whatever we want to power the data centers.
00:19:35
We need to build God like that's that's what they want.
00:19:40
And Eric, to your point, I, I think that is what they want.
00:19:42
And I feel like they've really got it.
00:19:44
Like, you know, I was, you know, commenting like all this point
00:19:47
about, you know, Trump being low energy and like, you know, the
00:19:49
Epstein story breaking while he's on stage, Like none of that
00:19:52
seems to really matter. They got the order, you know,
00:19:55
like it's, it's been the the path has been cleared.
00:19:57
It will matter electorally, but like they're willing to just
00:20:00
like, well, yeah, we won't talk about any of that.
00:20:02
Just like we'll we'll say you're awesome.
00:20:04
Like give us our clear cutting and like.
00:20:06
Let's go forward. Right.
00:20:08
The action plan is already out. The orders are already signed
00:20:10
like you know they're good to go.
00:20:12
Yeah, and they don't really need a Congress to support any of
00:20:15
these things anyway, so they might as well build as quickly
00:20:17
as they can until Trump, you know, completely loses his
00:20:19
ability to do any of the things that they want.
00:20:22
Because I I guess you can always sign E OS.
00:20:25
No one's really going to stop you from that.
00:20:27
And I don't even know how. I mean, is it clear actually
00:20:29
does you're talking to people there like whether or not these
00:20:32
executive orders were going to do all that much.
00:20:34
I mean, there's a lot of posturing with these things.
00:20:37
You're not actually changing that many rules.
00:20:39
They're not signing things into law.
00:20:40
You're not getting rid of active restrictions.
00:20:42
I mean, you know, this, we're poking around like a theme of
00:20:45
this administration is like, who cares what appropriators said
00:20:50
this money was for? Like, we'll do whatever we want.
00:20:52
So there is a degree to which, if the administration is willing
00:20:56
to say they want something and they're willing to bend all the
00:20:59
rules about using money appropriated by Congress however
00:21:02
they want, this is a time where executive orders are more
00:21:06
powerful than ever if they can navigate the bureaucracy.
00:21:10
What stood out to you among the EOS that you think is going to
00:21:13
be the most important? Because I have some thoughts,
00:21:15
but I'm curious as you were there, like what what mattered
00:21:18
to you? What stood out to me, there's
00:21:19
something that stood out from, you know, us as reporters
00:21:22
standpoint that could shape the form of AI, which I do think
00:21:25
the, you know, removing government contracts, unless
00:21:28
your AI isn't woke or whatever point will have impacts on how
00:21:33
this AI forms in the future. But I think from the business
00:21:36
standpoint, you know, streamlining regulations and
00:21:39
gutting climate rules to, you know, do this massive build up I
00:21:41
think will be the most consequential for the AI
00:21:43
industry. What do you think, Tom?
00:21:46
Yeah, I sorry, just back on like the, you know, government
00:21:49
contracts for AI. We we've seen a bunch of the big
00:21:52
companies, the the big labs and model makers sign, you know,
00:21:55
multi $100 million deals with the government recently.
00:21:58
Like I saw XAI got a big contract, which is interesting.
00:22:02
Open AI got one as well. So we're like at the beginning
00:22:05
stages of like actual tangible contracts between these
00:22:09
companies and and the government.
00:22:10
I do want to say as a point, it does often get misreported that
00:22:14
like opening, I signed a $200 million contract with the DoD.
00:22:17
That's not quite right. They signed like a $5
00:22:20
like basically exploratory contract with the potential of
00:22:23
it getting up to $200 million. So people should not report it
00:22:27
as like $200 million is going directly into open AI as like
00:22:30
some sort of government enterprise contract deal.
00:22:32
It's a little different than these things typically get
00:22:34
reported. But the one part that stuck out
00:22:37
to me, and again, I don't know how much this is going to matter
00:22:40
in the large scheme of things is the fact that, you know, and I
00:22:43
think Trump mentioned this in his speech, that AI model should
00:22:46
be given clearance to train on copyrighted materials.
00:22:51
And it's a huge issue right now that's playing out really
00:22:55
literally in the courts over, you know, the when all of these
00:22:59
models were built, you know, chat GPTS models, GPT 4,
00:23:02
whatever anthropics models, they basically ingested the entire
00:23:06
web and, you know, published materials in order to train on
00:23:09
these things. And as the New York Times is
00:23:11
currently suing right now, they New York Times, because it was
00:23:14
trained on their content, should be compensated for the
00:23:18
information that they created that was later used to create
00:23:21
these models, which are obviously now hyper competitive
00:23:24
with traffic to all of these new sites.
00:23:27
And I also saw there was, was it like a gal or was it like a Pew
00:23:29
poll or a Pew study today that showed that traffic has declined
00:23:33
substantially to a lot of new sites based on the AI overviews
00:23:38
that Google does. And so kind of like the doomsday
00:23:41
scenario that every news organization was thinking about
00:23:44
with the rise of AI and like summarization of articles may be
00:23:48
upon us because people are more willing to read an AI summary of
00:23:52
copyrighted material then then go to the article.
00:23:56
But I got super into this topic, like over the last couple days
00:23:59
and was talking to people about it.
00:24:01
Because if you talk to, you know, journalists or certainly
00:24:05
the lawyers who are representing The New York Times, they believe
00:24:07
like a grave injustice is being it's been done here right now
00:24:11
that that the fact that, you know, these models were trained
00:24:14
on New York Times copyrighted materials, like fucked
00:24:16
everything up for everyone. And they deserve money for it.
00:24:19
And so far, the most of these lawsuits have lost like Mehta
00:24:23
recently just won, you know, a case against this.
00:24:26
And it's a really fascinating conversation and like discussion
00:24:29
about what does it mean to consume and create material and,
00:24:35
you know, the, the, the, the history of the copyright.
00:24:38
And, you know, like when Google created its algorithm for its
00:24:41
search engine, it was also ingesting the entirety of the
00:24:44
web, creating cached pages of, of, of the Internet and creating
00:24:48
some sort of content. If you want to call a search
00:24:50
result content on top of that, that had specific effects on the
00:24:54
websites. And we are seeing now basically
00:24:58
again, this doomsday scenario play out where it is competitive
00:25:01
to a degree with the original content, whether or not they
00:25:06
were legally allowed to do it. And I mean, if we want to get
00:25:09
into a debate about it, I'm down to but like my stance as I spend
00:25:11
more time and it is like, yeah, I do believe they are legally
00:25:14
allowed to do it. And it and it comes down to this
00:25:17
question. And I think Trump kind of
00:25:18
touched on it, you know? Well, yeah, I was going to say
00:25:22
it. Can.
00:25:22
Can I say it? Yeah.
00:25:23
And when he was on stage, Trump kind of did a little when he
00:25:26
went off script, he did a little riff about this.
00:25:29
You can't be expected to have a successful AI program when every
00:25:32
single article, book, or anything else that you've read
00:25:35
or studied you're supposed to pay for, Gee, I read a book, I'm
00:25:39
supposed to pay somebody. And you know, we, we appreciate
00:25:42
that, but you just can't do it because it's not doable.
00:25:46
And if you're going to try and do that, you're not going to
00:25:48
have a successful program. I think most of the people in
00:25:50
the room know what I mean. When a person reads a book or an
00:25:54
article, you've gained great knowledge.
00:25:57
That does not mean that you're violating copyright laws or have
00:26:01
to make deals with every content provider.
00:26:05
And that's a big thing that you're working on right now.
00:26:08
I know, but you just can't do it.
00:26:09
China's not doing it. And if you're going to be
00:26:11
beating China, and right now we're leading China very
00:26:14
substantially and AI very, very substantially.
00:26:17
And nobody's seen the the amount of work that's going to be
00:26:20
bursting upon the scene. But you have to be able to play
00:26:25
by the same set of rules. So when you have something, when
00:26:29
you read something and when it goes into this vast intelligence
00:26:35
machine, we'll call it, you cannot expect to every time,
00:26:40
every single time say, oh, let's pay this one that much.
00:26:43
Let's pay this one just doesn't work that way.
00:26:45
Of course you can't copy or plagiarize an article, but if
00:26:49
you read an article and learn from it, we have to allow AI to
00:26:53
use that pool of knowledge without going through the
00:26:55
complexity of contract negotiations, of which there
00:26:58
would be thousands for every time we use AI.
00:27:03
He kind of touched on the point, you know, himself in a way that
00:27:06
I felt like was honestly kind of hard to go up against.
00:27:11
How do paywalls play into this? Like is there?
00:27:15
Is there just how I don't actually know the particulars?
00:27:18
Like how much is their argument that they're the information was
00:27:21
behind a paywall? I think if you know this gets
00:27:24
into like the robot dot TXT construction of websites which
00:27:29
basically allows the robots to scan the pages.
00:27:33
I, I'm pretty, I'm pretty sure stuff that's behind a paywall is
00:27:36
still probably scannable by by the robots and so that stuff
00:27:40
does like. The New York Times had set it up
00:27:43
so that it was scannable because they wanted search results.
00:27:45
Yeah, yeah, yeah. The same way that it can get
00:27:48
scanned by, you know, the web crawler also allowed these
00:27:52
models to be able to ingest the content on these pages.
00:27:54
The way someone framed it to me, which I was really interested
00:27:57
in, was like, the core of the question is really, do robots
00:28:00
have the right to read? Like, are they allowed to go to
00:28:03
these websites and read and learn things?
00:28:05
And is the thing that they created by reading it
00:28:07
competitive with it? Is it transformative or is it
00:28:09
derivative? If it's derivative, it's
00:28:11
competitive and it's in a violation of copyright.
00:28:13
If it's transformative, it's something new.
00:28:16
And I mean, these are the questions that are going to be
00:28:17
playing out probably in front of the Supreme Court.
00:28:20
But I should be allowed to say who can read read my stuff
00:28:24
right? And like if I have a paywall.
00:28:26
Would you be fine with your like newcomers stories not showing up
00:28:30
in a Google search because you don't want it to?
00:28:32
But why can't I draw a distinction and say I I'm OK
00:28:35
with Google and I'm not OK with you And like you're allowed to
00:28:37
do business with whoever you want.
00:28:39
Like I'm OK with businesses where they're they're trying to
00:28:42
link to me and get people to come to me and not OK with
00:28:45
businesses where they're trying to replace me and summarize me.
00:28:49
Like I don't know where to free market.
00:28:52
I can do business with who I want.
00:28:54
So so to me, I would sort of push more on this sort of like,
00:29:00
you know, should open AI have paid?
00:29:01
Yeah, paid to access the stories.
00:29:07
Well, people, you know, Anthropic did it in their own
00:29:09
way. You know, they bought literally
00:29:11
bought physical books and scanned them and then destroyed
00:29:14
the original material, which that seems like the nature of
00:29:18
the. Book, I mean a book, it's like,
00:29:19
yeah, you get it. You can do whatever you want.
00:29:21
You can, you know, consume it. Physical media you you have the
00:29:25
right to consume it once you own it.
00:29:27
I don't know how it's going to play out.
00:29:28
I mean, again, like every single lawsuit has has lost or or been
00:29:31
thrown out thus far. I think the New York Times
00:29:34
probably has the strongest 1 going right now just because
00:29:38
they're the New York Times. They probably have the best
00:29:39
lawyers to advocate for it. But I imagine this is going to
00:29:43
reach the Supreme Court. It like hurts my heart.
00:29:46
It's like a hard, like, on the one hand, you know, if AI is
00:29:51
going to succeed and AI is like clearly important, our models
00:29:54
need to be able to like scroll, you know, consume all this
00:29:58
information and like they're going to be of national
00:30:01
importance. And like China is clearly doing
00:30:03
the same. So just on a pragmatic level,
00:30:05
like clearly we need them to basically consume everything and
00:30:10
higher quality information. But I'm certainly sympathetic to
00:30:14
people who created all the valuable information saying, I,
00:30:18
I created this world, historically important thing.
00:30:22
Shouldn't I get money out of it? I do.
00:30:24
I do think a lot of the like the value in these AAI models, the
00:30:29
techniques, everybody's been able to copy the technique and
00:30:31
people who came up with the technique or some random
00:30:34
researcher, you know, a lot of the value in the models is the
00:30:37
actual underlying knowledge that's been accumulated by, you
00:30:42
know, a bunch of. Different companies.
00:30:43
My worry about this though is is that if it does get to a point
00:30:46
where you basically just have to compensate all the creators of
00:30:49
content to in order to use their material to create a model,
00:30:54
there are companies that can pay for it.
00:30:56
They're just giant tech companies and we're leading
00:30:59
towards a world where the only people that would be able to
00:31:02
build models are ones that have the 10s of billions of dollars
00:31:05
that you would maybe need to to pay off the creators to do it.
00:31:08
And that's fine. We're kind of already there
00:31:10
anyway, but it destroys the possibility of ever there, you
00:31:13
know, of open source models and, and I think obviously they're
00:31:17
not very big right now the open source community compared to the
00:31:20
proprietary's. But if you believe in that, and
00:31:22
I think a lot of people in the Internet world do have that as
00:31:25
their core belief that open source matters, this seems to
00:31:28
set an extremely high bar monetarily to build these sort
00:31:32
of things. And, and I think that's a bad
00:31:33
thing. Just to add on to that, in the
00:31:35
open AI action plan, you know that they were touting all day
00:31:38
yesterday, there was a provision around protecting open source
00:31:43
and a lot of founders that I spoke to said that that was a
00:31:46
really important win for little tech for them so that they could
00:31:49
continue operating. So.
00:31:50
Oh yeah, that that's a key thing.
00:31:52
I mean, honestly, in some of what Trump is doing is what he's
00:31:55
not doing, you know, it's like the Biden administration trying
00:31:58
to come up with rules and savior this a model this size we're
00:32:01
going to attract you. And now they're being expressly
00:32:04
pro open source. You know, there there was much
00:32:07
more it felt like in the bite administration, the anthropics
00:32:11
and opening eyes of the world. We're trying to sidle up to the
00:32:15
administration and say you can trust us part of the deep state
00:32:18
liberal machine. Don't trust the sort of Wild
00:32:21
West open source models. And clearly under Trump, it's
00:32:24
like, OK, open source is sort of that capitalistic, let everybody
00:32:29
compete spirit. And so that that's another part
00:32:33
of what they announced. Yeah.
00:32:35
So, so Eric, earlier you mentioned Dario's Modi, CEO, Co
00:32:38
founder of of Anthropic, also Cerebral Valley guest at one
00:32:43
point, all these things are and more on his LinkedIn, he
00:32:47
recently sent an internal memo to the staff at Anthropic saying
00:32:51
that they are planning to or in discussions to take some money
00:32:54
from the Middle East. Where was it specifically?
00:32:57
Was it was it Qatar or? Qatar definitely is one of the
00:33:00
major sources. I forget if you called out a
00:33:02
specific country. Yeah, yeah.
00:33:04
And, you know, it seemed like he was a little sheepish about the
00:33:07
whole thing, or at least defensively explaining it,
00:33:09
saying, listen, we did not intend to take money from the
00:33:12
middle. East, he literally says like,
00:33:13
you can't just have good people benefit from your success if you
00:33:17
want to be a big company. Yeah, yeah.
00:33:20
He's teaching us about the way the world works.
00:33:21
He basically gives us the Ned Betty speech in in from network
00:33:25
about how the world works and how you need to kind of accept,
00:33:28
you know, the, the, the primordial powers of of the
00:33:30
universe. It is like capitalism sort of at
00:33:33
its essence. You know, I like capitalism, but
00:33:36
there is a certain like, oh man, like our job is to do this.
00:33:40
You look at the trade off. You know, we're gonna we need to
00:33:43
raise money. Money is money.
00:33:44
Like I, I'm. I mean, this is just one of many
00:33:48
compromises that Dario has had to make as the CEO of Anthropic.
00:33:52
And if you remember the history of this company, he broke off
00:33:55
from Open AI because he was pissed off about taking money
00:33:59
from Microsoft and it kind of polluting this effective
00:34:01
altruist mission of, of the original company and, and the
00:34:05
nonprofit. And then he goes off and lo and
00:34:08
behold, there's a shit load of money that can be made.
00:34:10
And also it cost a lot of money to make these models.
00:34:12
And, you know, he goes off and raises as much as almost any
00:34:15
other company other than Open AI and now realizes that, like,
00:34:19
when you need to keep raising this level of money, there's
00:34:21
really only one place you can do it, and that's the Middle East.
00:34:24
And so the guy is like pretty experienced at compromises.
00:34:27
And it's funny that he's even like, trying to, you know,
00:34:30
explain it away when it's like. I mean, I don't have any
00:34:33
objection to him taking money in the Middle East.
00:34:35
I think the sort of capitulation is interesting to interrogate.
00:34:40
To me, it's interesting right now, just like thinking about
00:34:44
Trump's executive order, right? If we are counting on companies
00:34:49
to have backbone and say, I guess we won't sell the
00:34:51
government, we're going to have to sell out our values on, you
00:34:55
know, thinking about what the results the model produces.
00:35:00
This is not a sign that the the companies will keep their
00:35:02
backbone, right? They're like, well, our job is
00:35:05
to chase the almighty dollar. And yeah, the government is an
00:35:08
important contractor and this is a small capitulation about a
00:35:12
topic we don't that is not central the model like whatever
00:35:15
the money sort of swamps the principle.
00:35:18
And so I, I do think, you know, it's the classic like you give
00:35:21
up one principle and slowly you find yourself giving them all
00:35:26
up. And I don't think this is
00:35:28
necessarily the line that anthropic needs to hold on to.
00:35:32
And I still generally trust them as a moral decision maker.
00:35:37
But you know it. It's small capitulations on the
00:35:41
way to to giving up your morality.
00:35:44
It feels like that was kind of the long arc of open AI, right?
00:35:47
Like, you know, as the an AI that is benevolent to mankind.
00:35:51
And then, you know, there were moral conflicts between this,
00:35:53
but. Open AI was always run by one of
00:35:56
the most power hungry men alive, so I don't know that that's less
00:36:00
surprise. I they yes, they became they
00:36:03
started a non profit. But you know, Sam was always
00:36:05
extremely ambitious. That's that's sort of the that's
00:36:08
line of open AI. That's very fair.
00:36:10
That's a good point. But all this just to say, I'm,
00:36:13
I'm saying even the effect of altruist AI company goes the way
00:36:18
of capitalist enterprise, so. Yeah.
00:36:22
And listen, like if there are going to be paying customers
00:36:24
that want to buy the ability to make biological weapons with
00:36:27
your AI, do you want to win or not like?
00:36:31
They probably pay really well. Yeah, like it's and that always
00:36:35
is the thing. By the way, I've noticed that
00:36:36
there's like fears about the usage of of of large language
00:36:39
models to make bio weapons. And, you know, to date, Dario
00:36:43
remains against using this technology to do that.
00:36:46
But listen, we make one compromise.
00:36:49
You make a lot of compromises. Well.
00:36:50
I was going to say there's always that fear on the bio
00:36:52
weapons, but the the qualms around other weapons and, you
00:36:55
know, partnering with the Defense Department, those are
00:36:57
gone. So yeah, it's just the bio
00:36:59
weapons. I'm interested in like, you
00:37:02
know, the value of a corporate morals and and strategy as to
00:37:09
you know, like, is it about recruitment?
00:37:10
Is it about winning over business or is it about just
00:37:12
sleeping at night? Well, Anthropic has always had
00:37:16
this position that on AI safety they would create a race to the
00:37:20
top. And you could argue that there's
00:37:23
some truth to this. Like the idea was we'll have
00:37:26
scruples and therefore engineers with scruples will come here and
00:37:31
that will create pressure on other companies to, to have
00:37:35
scruples themselves. And I, I do think we've seen way
00:37:38
fewer defections from Anthropic than we've seen at Open AI.
00:37:42
On the other hand, you know, I don't think, Mike, I don't think
00:37:46
Meta is shouting about its scruples.
00:37:48
It's shouting about the power of the American dollar.
00:37:50
And so I don't know that Anthropic's race to the top is
00:37:54
necessarily trickling out to the rest of the AI industry.
00:37:58
But I do think it's had this sort of view that having sort of
00:38:02
principles about we're really going to test these models,
00:38:04
we're going to have robust safety, would help it win
00:38:08
because it would attract, you know, these AI engineers are
00:38:11
getting paid more money than they could ever use.
00:38:13
So to some degree, they are the type of people that will decide
00:38:17
where they work based on other things besides just money.
00:38:20
And you know, having sort of principles is one of them.
00:38:24
I mean, that's the missionary versus mercenary, uh, breakdown.
00:38:27
And it was funny for Sam to be the one that kind of put that
00:38:31
into, you know, the conversation and into the discourse.
00:38:34
I mean, I he's a missionary, but it feels so ambitious, you know,
00:38:38
it's like, it's like I want to be the God king of, you know,
00:38:40
AI. Well, if you're if you're a
00:38:42
missionary surrounded by mercenaries too, like you can,
00:38:45
you know, I don't know how much longer that's going to last for
00:38:48
you if it's really ultimately A mercenary world about attracting
00:38:52
talent. And I think everyone does really
00:38:54
have a number, whatever. I don't want to go deep into
00:38:56
like the talent wars again, but. I mean, you've talked to some of
00:38:59
these people. Yeah.
00:39:00
Well, I think they they are they everyone has a number.
00:39:06
Yeah, I don't understand the psychology of the person who's
00:39:08
like, I make $10 million in four years, Like now I need to make
00:39:14
75 or whatever. I guess I Yeah, that that person
00:39:18
is distant from me. What?
00:39:19
What is the thing that they want?
00:39:21
Well, I think it's just the human impulse of like, I make
00:39:23
$10 million and this guy who's as talented as I am, or maybe
00:39:27
even less talented I am, is making 75.
00:39:29
Like, fuck you. Like, why do you get a third
00:39:31
house? Like that just doesn't that
00:39:33
doesn't sit well with me. So I'm going to like if that
00:39:36
that fucking money's out there, I would be stupid not to take.
00:39:39
I think that's a much more powerful.
00:39:40
Argument and, you know, you start looking at all the
00:39:42
different like marble countertops that you could get
00:39:45
and you're like, well, you know, it adds up, you know.
00:39:48
There's always a place to spend the money.
00:39:49
That's never the issue. All right?
00:39:52
Anyway, good luck, Dario, with that Middle East money.
00:39:55
We'll, we'll, we'll be watching. For it, they still invest, you
00:39:58
know, like they they read this memo where they're like, yeah,
00:40:01
it's distasteful, but we'll do it.
00:40:04
Right, we're about to take. Money from some of the most
00:40:06
disgusting. People in the world.
00:40:08
This is going to be the worst of both worlds.
00:40:10
You know, they they make the moral capitulation, then they
00:40:13
don't get the money because they stuck their nose up at it.
00:40:16
Yeah, well, capitalism runs both ways too.
00:40:18
Like they they want to invest in these things because they want
00:40:20
to make money as well. So you're right.
00:40:22
Yeah. Like they're not just.
00:40:25
All. Right, that that about does it
00:40:27
for us this week. Thanks so much for listening and
00:40:30
we will catch you back here next week.
00:40:33
Subscribe to newcomer.co and we'll see you next week.