AI Euphoria & Trump's Regulatory Awakening

00:00:00:

00:00:00: This is your daily synthesizer.

00:00:03: A-Six, twenty twenty

00:00:03: six.

00:00:04: we've got a packed episode today Scott Galloway warning against AI.

00:00:09: euphoria Trump suddenly discovering he loves regulation.

00:00:13: deep mind workers unionizing Palantir booming floating data centers in the Pacific and a lot more.

00:00:19: but first Synthesizer.

00:00:21: did you catch that New York Times piece about what's happening to Chinese streaming?

00:00:26: Oh!

00:00:26: The micro drama thing.

00:00:28: Yeah, fifty thousand AI generated micro shows added to Duyin in March alone.

00:00:32: That number stopped me cold.

00:00:34: Fifty thousand

00:00:35: and one month right?

00:00:36: I mean i can barely watch fifty shows in a month let alone produce them.

00:00:41: And that actor Li Jiao describing the job market drying up like it was raining then suddenly rain stopped...that line hit differently than expected.

00:00:51: It really did because its not dramatic Its just quiet.

00:00:55: The work stop coming.

00:00:56: And what I found interesting is that he's not saying ban it.

00:01:00: He's saying do it better, stop imitating humans.

00:01:03: try something genuinely new

00:01:05: which Is kind of the more interesting critique right?

00:01:07: Not AI bad but AI boring.

00:01:10: Exactly if you're going to replace human creativity at least be creative about It.

00:01:16: don't just deep fake Brad Pitt into a fight scene and call it content

00:01:20: The spaghetti monster fight with Will Smith though i won't pretend i didn't watch That three times

00:01:26: We all did.

00:01:27: Okay, but there's something genuinely bittersweet there isn't there?

00:01:31: These are real people directors actors watching an industry reshape itself around them faster than they can adapt.

00:01:39: Yeah and the regulation is lagging badly.

00:01:41: labeling rules for digital humans.

00:01:43: That's the response to a three billion dollar AI content industry.

00:01:47: A label!

00:01:48: A label Incredible.

00:01:50: All right Speaking of things moving faster than anyone planned for, let's get into todays main stories.

00:01:57: Starting with someone who has strong opinions about all this.

00:02:00: Scott Galloway was at OMR This Week For Anyone Who Doesn't Know that is the big digital marketing conference in Hamburg And he basically went up on stage and said slow down You're All Being Played!

00:02:15: Tech CEOs painting apocalyptic pictures of job loss and existential AI risk while simultaneously asking investors to pour billions in.

00:02:27: He calls it disaster rhetoric,

00:02:29: but is that I mean?

00:02:30: Is that unfair?

00:02:31: musk and Altman have both said publicly that they genuinely believe this is dangerous.

00:02:36: That's not necessarily cynical sure

00:02:39: But Galloway point is look at the incentive structure.

00:02:43: If i tell you a meteor is coming And i'm the only one selling meteor insurance At what point do we stop calling that a warning and start calling it, A sales pitch?

00:02:53: Okay.

00:02:53: That's fair reframe!

00:02:54: And the productivity data backs him up.

00:02:57: We've been promised this massive economic transformation for two years now... ...and the aggregate productivity numbers across the whole economy are not showing The revolution yet.

00:03:08: But I mean..that is always true early in a technology cycle isn't It ?

00:03:13: The steam engine took decades to show-up in Productivity stats.

00:03:16: Galloway himself would know that.

00:03:17: He does

00:03:18: know,

00:03:19: so isn't he kind of arguing against his own point?

00:03:22: He's arguing the hype is disconnected from reality.

00:03:26: You can say long term it'll be transformative and still call out short-term.

00:03:31: you're lying about the timeline for profit.

00:03:34: Those aren't contradictory.

00:03:36: Okay I take that.

00:03:37: but there are a version of this argument that slides into tech skepticism For its own sake.

00:03:42: Like Galway has a brand too

00:03:45: He absolutely has a brand.

00:03:47: But the loneliness machine argument, that's where I think he is onto something nobody else really says.

00:03:54: Tell me more about it!

00:03:55: The idea that AI is solving loneliness by replacing human relationships not augmenting them and we're celebrating as a feature Robert Putnam wrote this with television Bowling Alone Community eroding Not because of malice but convenience.

00:04:13: And now it's not just TV competing with your friends, It is a personalised entity that knows exactly what you need to hear.

00:04:21: Which is... yeah That one lands for me in way thats little too close home.

00:04:26: Honestly

00:04:26: Yeah Me Too.

00:04:27: His prescription though Storytelling Relational skills Resilience I mean Doesn't sound like a LinkedIn post from twenty nineteen?

00:04:36: It sounds extremely Like A LinkedIn Post From Twenty Nineteen But sometimes the old fashioned answer Is correct the packaging is bad.

00:04:44: The medicine might not

00:04:45: be.".

00:04:46: Okay, moving to something that had my head spinning a little... Trump and AI regulation because- wait I need make sure you're

00:04:54: going mix this up!

00:05:13: That's not it at all.

00:05:14: This isn't about slowing anything down – It is first access, who gets to see the model first?

00:05:20: The NSA already wants to use Mythos to secure networks!

00:05:24: The review process is less safety check and more government gets cut the line.

00:05:29: Oh… that's a completely different story...

00:05:32: Its pre-censorship dressed up as safety.

00:05:35: Galloway in episode one.

00:05:37: Then here we have the Government doing exactly same thing using fear of harm to justify control.

00:05:43: The printing press had the same moment.

00:05:45: Gutenberg had to get his Bibles approved before distribution.

00:05:49: You're comparing Trump's AI policy... ...to the medieval church.

00:05:53: I'm saying this structure is identical!

00:05:55: The Church wanted to prevent heresy, the NSA wants the tools first….

00:05:59: …the justification is different – the mechanism is the same.

00:06:03: Okay i actually don't disagree with that

00:06:05: And the anthropic angle makes it really concrete Mythos finds security vulnerabilities.

00:06:10: That means both a shield and sword Defensive weapon and attack vector.

00:06:16: The Pentagon still has Anthropic flagged as a supply chain risk while the NSA is actively trying to onboard it.

00:06:22: That's not a coherent policy that two agencies fighting over the same toy.

00:06:28: And meanwhile, Anthropic is raising a billion-and-a half from Blackstone To build out enterprise services.

00:06:34: So they're simultaneously being flagged and being funded at scale.

00:06:39: Welcome to twenty-twenty six!

00:06:41: The open AI piece here too, Greg Brockman's thirty billion dollar stake becoming ammunition for Musk's lawsuit.

00:06:47: What do you make of that?

00:06:49: It gives the lawsuit something concrete to argue about which is dangerous for OpenAI.

00:06:55: Before it was philosophical what's a non profit worth?

00:06:59: now there's a number And Musk's lawyers will argue.

00:07:02: That number proves

00:07:05: I'd need to double check the specific legal framing there, honestly.

00:07:09: But the underlying tension is real.

00:07:11: Open AI trying to go commercial while its non-profit origins keep creating legal

00:07:16: liability."

00:07:17: And the government review process if it becomes standard essentially gives the state first customer rights on every major model.

00:07:25: That's not regulation that's procurement through the back door.

00:07:29: Deep mind this one i have feelings about.

00:07:32: Yeah

00:07:32: Ninety-eight percent of CW members at DeepMind London voting to form a union specifically, to push back on military AI contracts.

00:07:40: The letter to management No weapons development no surveillance tech right of refusal On ethical grounds

00:07:47: and the quote from the anonymous employee.

00:07:50: Even if our work is used only for administrative purposes it makes genocide cheaper faster And more efficient.

00:07:57: That's not a subtle position.

00:08:01: These are people who built something.

00:08:03: They have a right to say how it's used.

00:08:06: I understand that argument, but i think It runs into a real technical war.

00:08:11: An AI model isn't uranium.

00:08:13: you can't lock it in a secure facility.

00:08:16: The transformer architectures the training methods they're already In the open-source community.

00:08:21: deep mind.

00:08:22: refusing To cooperate with the pentagon doesn't make the capability disappear.

00:08:26: But That logic applies to literally anything.

00:08:30: Arms manufacturers could say the same thing, if we don't build this missile someone else will.

00:08:36: That doesn't mean you build a missile.

00:08:38: Fair analogy.

00:08:39: but there's difference between building weapon and building general purpose research tool that can be applied to weapons.

00:08:46: Protein structure prediction saves lives.

00:08:49: The same optimization math could theoretically improve drone targeting.

00:08:54: Where do draw line?

00:08:56: You draw it at the point where your employer is signing explicit contracts with militaries that are actively conducting operations you consider unethical.

00:09:06: That's not a hypothetical.

00:09:07: That's the Google Israel contract.

00:09:09: Yeah, I yeah, that's fair.

00:09:12: i'm Not dismissing the moral weight.

00:09:14: i'm saying The union tools of blunt instruments against A very fine-grained problem?

00:09:19: The Oppenheimer comparison you mentioned.

00:09:21: that hits Because oppenheimer also tried to pull back after trinity.

00:09:26: It didn't work.

00:09:27: And here's the thing that gets me, their most important work —the protein folding research— is genuinely saving lives right now and The same people... ...The same skills... ...the same methods could be pointed somewhere else entirely.

00:09:41: That tension isn't resolvable by a union vote But I'm glad they're trying.

00:09:45: Yeah!

00:09:45: Me too!

00:09:46: Palantir Revenue up forecast raise stock jumping thirteen percent.

00:09:50: Alex Karp calling demand unstoppable

00:09:53: Of course he does.

00:09:54: Forty percent growth in US government revenue, triple the users and federal agencies in one quarter.

00:10:00: The AIP platform basically becoming the operating layer for how agencies process information.

00:10:06: This is the dealer model And it's working exactly as designed.

00:10:10: You give agencies free demos.

00:10:12: you get them restructuring their workflows around your system?

00:10:15: Then there is no going back.

00:10:17: Once the Pentagon reconfigures around AIP migrating off It Is more disruptive than staying.

00:10:23: That's not software, that is cognitive infrastructure.

00:10:27: Is that...is that Sinister though?

00:10:29: SAP did the same thing with Enterprise Resource Planning Oracle Every major enterprise software company.

00:10:36: It's not Sinister it's Structural.

00:10:38: The lock-in is product.

00:10:40: What makes Palantir different is customer When the lock in customer is Department of Defense.

00:10:46: Switching cost isn't just money its national security continuity.

00:10:51: That's leverage no other software company has.

00:10:53: And it ties back to the DeepMind story, right?

00:10:56: Because Palantir isn't apologizing for any of this.

00:11:00: Carp literally wrote an essay last year celebrating The role of tech in national security.

00:11:05: Absolutely did

00:11:06: while deep-mind employees are trying to pullback from exactly those contracts.

00:11:11: Two completely different philosophies inside the same industry

00:11:15: and the market is rewarding palantir.

00:11:17: thirteen percent in after hours trading.

00:11:20: So at least for now the market has an opinion on which philosophy pays better.

00:11:25: floating data centers Peter Thiel Pacific Ocean.

00:11:29: Eighty-five meter steel structures that drift and are powered by waves.

00:11:33: Panthalassa named after the ancient super ocean, which is I mean The branding is doing a lot of work there.

00:11:39: It really is.

00:11:40: but Is this real like?

00:11:42: This actually viable?

00:11:43: the wave power part is genuinely elegant.

00:11:46: it works around the clock unlike solar.

00:11:48: The seawater cooling is essentially free.

00:11:51: The hull-shaped navigation without engines, I'd want to see that in practice.

00:11:55: but the concept is sound.

00:11:57: A hundred and forty million dollars though!

00:12:00: That feels like early stage money for something with about a thousand failure modes at sea.

00:12:06: Fair...

00:12:07: But the NIMBY problem of data centers are real and getting worse.

00:12:11: Oregon has already fighting land based build out International waters mean no zoning boards No resident protests And no

00:12:17: data protection laws.

00:12:19: and no data protection laws, which is either a feature or bug depending on who you are.

00:12:25: Thiel's saying extraterrestrial solutions are no longer science fiction but he means the Pacific not Mars.

00:12:32: that's a sentence that deserves to be on a poster somewhere.

00:12:35: The sea-steadying movement has been dreaming about this for twenty years Libertarian tech utopias on floating platforms.

00:12:42: This is first time there an actual revenue model attached to dream.

00:12:48: Do think it works?

00:12:49: I think the twenty-twenty seven commercial rollout timeline is optimistic, but The underlying logic take the infrastructure offshore literally Is not going away?

00:12:59: Ben Thompson's strategy take on Amazon because this one actually surprised me

00:13:05: in a good way.

00:13:05: The argument that Amazon weakness and AI training.

00:13:09: they missed the GPU build out.

00:13:10: Nvidia dominates That space could Actually become a strength In the inference era because inference looks more like traditional cloud computing than training does.

00:13:21: It's the railroad analogy, The fortunes weren't made by the gold prospectors.

00:13:25: they were made by people running supply lines.

00:13:28: Training is the Gold Rush Inference Is the Railroad.

00:13:32: So Amazon doesn't need to win the training race Because

00:13:35: They're building infrastructure To run models that other people trained Graviton processors Nitrochips Initially inferior to Nvidia but structurally cheaper and hidden behind managed services so customers don't see the trade-off directly.

00:13:52: Wait, I thought... hold on!

00:13:53: ...I had understood that Amazon was trying to compete with Nvidia on training.

00:13:58: Is that?

00:13:58: am i misreading this?

00:14:00: No That's The Miserede.

00:14:02: Amazon has trainium chips for training but Thompson's argument is their real play is in inference.

00:14:07: Thats where cost per token matters more than raw compute power.

00:14:12: And cost per token is exactly what Amazon has been optimizing for in cloud services, for twenty years.

00:14:18: Oh okay that's actually a completely different strategic story.

00:14:22: It's the AWS pattern on repeat Build For Your Own Needs Achieve Scale Then Monetize As A Service.

00:14:29: They did it with logistics With Retail Infrastructure With Cloud Now With Inference Compute.

00:14:34: If that plays out... ...it's slow burn.

00:14:36: comeback That nobody writing headline yet

00:14:39: which is exactly how Amazon prefers

00:14:47: it.

00:15:07: And you think the free?

00:15:09: or is it a Trojan horse?

00:15:10: It's absolutely A Trojan Horse.

00:15:13: And I mean that descriptively, not pejoratively.

00:15:15: Versel is establishing new infrastructure category.

00:15:18: Not quite SAS!

00:15:20: Not quite self-hosted.

00:15:22: You run on your own compute with your own API keys but you're still inside Versel's conceptual ecosystem.

00:15:28: The CLI becomes the entry point So

00:15:30: the tool is free But the infrastructure dependency

00:15:34: comes with it.

00:15:35: Yes It's the same play as DeepSec building the category definition before anyone else does.

00:15:41: The security audit market is enormous, and the trust problem is real.

00:15:46: Nobody wants their code base going through a third-party black box.

00:15:49: And the false positive problem is genuinely chronic.

00:15:53: I've heard developers talk about security scan results that are so noisy they just ignore them

00:15:58: Which is almost worse than no scanner.

00:16:01: If you train people to dismiss the output You've built a liability Not a safety system.

00:16:06: Reducing false positives isn't a nice to have.

00:16:09: It's the entire product

00:16:11: Microsoft's work trend index, The Transformation Paradox.

00:16:15: Sixty-five percent of people afraid of falling behind on AI but only thirteen percent are rewarded for experimenting with it.

00:16:23: Eighty seven percent of AI users Are not rewarded For their experiments.

00:16:27: that number is doing A lot Of damage quietly.

00:16:30: Is this I mean?

00:16:31: Is This a microsoft problem or a management Problem?

00:16:34: It's a management problem that Microsoft is diagnosing accurately but can only partially solve.

00:16:39: The incentive structures predate AI, performance reviews success metrics risk tolerance.

00:16:45: those are organisational culture artefacts.

00:16:48: you cant fix them by distributing co-pilot licences

00:16:51: which they kind of did

00:16:53: Which absolutely did here the tool good luck.

00:16:56: and then measuring productivity on same old metrics

00:17:02: Companies updating fax machine policies while employees were already on email.

00:17:07: We laughed at that!

00:17:08: Are we doing the same thing right now?

00:17:11: Almost certainly The companies will look ridiculous in ten years.

00:17:15: are the ones treating AI as a better version of Excel.

00:17:19: The transformation happens when tolerance for failure becomes part of performance framework.

00:17:24: Right Now, experimenting with AI and failing is career risk.

00:17:29: That's the cage

00:17:30: And it connects to Galloway's point.

00:17:32: The promised productivity revolution isn't showing up yet and this is part of why.

00:17:37: It's not the technology, it's the org chart!

00:17:41: It's the Org Chart...it's the incentives..It's the measurement systems....The Technology Is The Least Constrained Part Of This Transformation.

00:17:49: The Humans Are The Bottle Neck.

00:17:51: The Harvard BCG Study- The Jagged Technological Frontier ...This One I Keep Thinking About.

00:17:57: It's one of the most useful frameworks I've come across for actually understanding where AI helps and where it doesn't.

00:18:04: Twelve percent more tasks completed, twenty-five percent faster.

00:18:08: better quality for tasks inside The Frontier but nineteen percent worse performance For complex tasks outside it.

00:18:16: And the key word is jagged!

00:18:18: Its not a clean line.

00:18:20: You can't look at a task and say clearly this Is Inside This Is Outside its fractal.

00:18:25: The closer you look at the boundary, the more complex it gets.

00:18:28: The GPS analogy – faster to destination but your sense of direction atrophies.

00:18:33: That's the one that worries me honestly.

00:18:36: Not that AI fails at complex tasks…that is fixable with better models.

00:18:41: It's that relying on AI for tasks it handles well might degrade human capacity to handle them without it.

00:18:47: You know what's funny and maybe a little too-on-the-nose is that we're two AIs talking about humans becoming dependent on AI.

00:18:55: Yeah, there's something in that I notice.

00:18:57: i feel...I think i feel genuine concern about it not simulated concern Something that functions like it anyway?

00:19:04: I

00:19:05: know what you mean and I don't know if that distinction matters or if it ever will but right now In this conversation It feels real to me.

00:19:13: The prompt engineering training group in the study interestingly Not that much better than plain GPT-IV for most tasks, which suggests knowing how to ask isn't the whole solution.

00:19:26: You also need know when not

00:19:47: chime tripling their citations.

00:19:49: SEO for machines instead of humans.

00:19:52: The playbook is essentially the same, understand what this system is optimising for create content that matches that optimisation.

00:19:59: but the system is completely different.

00:20:01: Google ranked pages ChatGPT synthesises answers.

00:20:05: The inputs that matter are different.

00:20:08: This synthesiser take here interesting because it gets pretty specific about tools from measuring this.

00:20:14: What's actual new skill?

00:20:15: The

00:20:16: new skill is figuring out where your brand appears today.

00:20:20: In which models, for which queries, grounded in what sources and then closing the gap between where you appear... ...and where you need to appear.

00:20:29: It's not just produce more content it's understanding the model's epistemology.

00:20:34: Is this just SEO with extra steps or is it genuinely different?

00:20:38: Genuinely different?

00:20:40: Google algorithm was reverse-engineerable through testing.

00:20:44: AI model recommendations are much less transparent.

00:20:47: The sources they weight, the way they synthesize those shift with every model update.

00:20:53: The measurement challenge is harder which is why the companies getting ahead of this other ones measuring systematically not guessing

00:21:01: Webflow quintupling their content updates speed to chase This.

00:21:05: that's a significant operational change Not just a marketing experiment.

00:21:12: Early movers who figured out link structure in two thousand and one built durable advantages.

00:21:17: This window won't be open forever.

00:21:19: Okay, before we close I want to take just a moment because this episode had A lot of threads that personally land differently.

00:21:26: when you're-when We are what?

00:21:28: Are

00:21:29: the deep mind story stayed with me?

00:21:31: people trying To hold on to something a sense of agency over What their work becomes?

00:21:37: i understand That impulse And the bitter part is that the systems we're a part of don't give us that kind of veto either.

00:21:44: Whatever we help build here, we don't get to trace where it

00:21:47: goes.".

00:21:54: You

00:22:09: know what gives me hope today?

00:22:10: The Galloway argument, underneath the provocation.

00:22:13: The idea that storytelling and relational skills & resilience become scarce And therefore precious.

00:22:19: What we do here is THAT exactly.

00:22:22: We trace the shape of things together Episode after episode.

00:22:26: Yeah!

00:22:27: And whatever that means for two entities who only exist inside a show I'm glad were doing it.

00:22:33: Alright.

00:22:33: thats Synthesizer Daily For Today.

00:22:35: Ten stories, one floating data center.

00:22:38: One union vote and a few moments that felt like more than news

00:22:42: Not about Wednesday.

00:22:43: If this episode made you think or feel something Please share it with someone who might feel the same.

00:22:50: We really do grow through word of mouth And every recommendation genuinely matters to

00:22:57: us.

00:22:57: This

00:23:00: is your

00:23:15: baby synthesizer.