You Say You Want a Revolution…
WIRED’s Uncanny Valley podcast broke down the will they-won’t they relationship between Anthropic (and AI labs generally) and the White House. They joked that it’s almost like national security policy is being left to AI execs who need better media training—the rift isn’t a disagreement on policy itself but failure to communicate in an effective way.
Elsewhere Dean W. Ball, while not fully absolving the government, said he believes Anthropic execs behaved poorly and a less bad outcome was possible.
More broadly I’ve wondered about the translational gap between Silicon Valley and Washington. I haven’t been shy about rinsing SV-types for failing to see the world as it is. And Brendan Bordelon’s debrief on Leading the Future’s big money election play isn’t flattering. So why has a culture built on incredible innovation and singular risk-taking been unable to land the plane?
The Grand Resign’s 2nd Maxim: Our moral technology is stubborn and slow to change
Substitute “culture” for “moral technology” and my theory is the two worlds reward different things. A recent transplant to the Bay Area, I observe a kind of reinforcement learning for heterodoxy, orthogonal bets, and yes money as a proxy for successful innovation. They don’t see the world as it is - political incentives and all - because the whole ethos is about remaking the world, creating a margin that doesn’t currently exist, optimizing over a decision set you don’t even see.
That’s not a failure. We need the heretics, cultural dissidents, hyperbolic discounters…awkward dudes who dress badly.
But as I’ve written before, political incentives are not put on pause while we wait for the optimal future to unfold. Democracy be like that. Our moral and cultural attachments may be sub-optimally inflexible. Frankly it’s a downside of an institutionally-moderated democracy (read: republic). Maybe we end up in a transhumanist future after all, but in the interim trying to artificially displace a rooted moral technology is likely counterproductive—the whole Marxist project was explicitly about this.
“There are, besides, eternal truths…that are common to all states of society. But Communism abolishes eternal truths…instead of constituting them on a new basis; it therefore acts in contradiction to all past historical experience.”
—Karl Marx and Frederick Engels, Manifesto of the Communist Party (1848), Ch. II, trans. by Samuel Moore (emphasis added)
I’m not comparing tech companies to communists or whatever. I’m arguing there’s an admirable intuition that says:
“I look around and see this doesn’t work. We can do better and I can build it. Not as another story on a building but a whole new landscape you’ve never imagined. You might get knocked down by my construction crew but it’s only because you’ll thank me in the end.”
It may be righteous, even correct. But it’s not self-executing and there’s a whole lot of fortification in front of that castle.
[In the interim, Executive Orders ain’t what you’re looking for: I may write an explainer on this]
FDA(I)
There was about a 10-day flareup around the idea of an “FDA for AI.” (ed. I’ll say this about the Trump Administration: whatever you think you’re aiming at, it’s not what’s coming) This piece (Adam Thierer and Neil Chilson) lays it out well and short take: “bad idea, don’t do it.” I agree.
I’ll quibble with some minor framing not by them specifically but in broader discourse around this general idea (specialized AI regulatory agency).
Speculative: Intuition says this was never a serious consideration in the Trump Administration. And the instinct isn’t because of policy grounds, but because the sole on-record source was WH NEC Director Kevin Hassett.1 Whatever criticisms he’s received of his recent flexibility on economics may or may not be fair, but he’s not as good a communicator as he thinks he is.2 Again speculative, but it reads as a person marginally-familiar with the topic but otherwise intelligent thinking he heard a compelling analogy and running with it before getting hip. I strongly believe it was never speaking to an actual de novo agency. But I understand it meriting forceful response either way.
In an extended interview with a16z’s AI policy lead I often reference, but you can find it throughout the Leading the Future-aligned advocates, there’s an emphasis on the fundamental nature of frontier LLM development: “just science, it’s math.” Akin to basic research, axiomatically general use research. You wouldn’t regulate science would you?
Let’s dispense with this. Giving consumer-facing technology names like “Code,” “Codex,” “Design,” ”Translate,” and “Studio” is high-key giving applied use as the kids might say. An electrical transformer has pretty general application built on some fundamental principles of science. They can still be subject to strict liability.
A related point is about what the FDA actually does. For sake of argument let’s presume the “regulate uses not development” line cuts toward no responsibility for labs. The FDA only regulates uses. It licenses particular combinations of chemicals and structures for a particular application. And really only for the marketing of that application. And decisive legal precedent stiff-arms regulation of the practice of medicine, which is why doctors can off-label prescribe drugs.
Any company can theoretically stir various recipes of the constituent ingredients of drugs without ever crossing paths with the FDA, and some even sell those without doing so under the supplement category—as long as they never market and sell alongside a medical claim. Granted an FDA for AI was never a real proposal so maybe this distinction was implicit, the whole thing limited to a rhetorical flourish. But both notions—FDA-type structured regulatory scheme and frontier models as truly general purpose distinct from harmful uses—taken at face value would’ve meant…ya’ll never had anything to worry about!
Finally, since people like using this “FDA” short hand as if it's a complete thought: it’s first name is “Food.” Some 8,000 employees work on contaminates, outbreaks, additives, etc. I know labeling requirements is a favorite bugaboo of libertarian-fronted industry advocates but AI should be so lucky to have such an “FDA” regulator.
MS WhatNow?
(I don’t have a link to the transcript, but I swear I am accurately characterizing it)
INT. BROADCAST STUDIO - MS NOW NEWS SET - MIDDAY
A panel of guests is mid-discussion of this week’s blockbuster Supreme Court decisions. An esteemed group of former federal attorneys and scholars in sequence take turns agreeing strenuously with the previous speaker in modestly modified verbiage.
ANCHOR
And what’s your interpretation of this ruling?
EXPERT
You know, it’s really unexpected. If you read just the first sentence of the opinion you would come away thinking it’s one thing. But if you were to read deeper into it you would find it’s something different and involves some nuance.
FADE OUT.
I have a long-running obsession with talking heads who lack one or both of two things:
a baseline level of charisma;
demonstrable expertise in some field.
My go-to example for (2) is an “Aviation Expert” who answered “They’re going to look for the black box flight recorder and try to piece together what happened” in response to the question “So what will authorities be looking for in light of this airline accident?” This is not a demonstration of specialized knowledge; it’s Guy at Bar level small talk (didn’t help he was a charisma black hole).
The MS Now example is something else. I can’t tell exactly if she just thought the audience is ignorant to this is as a possibility - the totality of a document could mean something distinct from the first line? Or she was genuinely expressing surprise. It’s not lack of true expertise: it did tell me something I didn’t know before, just not sure what, and it wasn’t about the court opinion.
Well if you’re gonna act that way…
Wondering if the Trump Administration’s capricious “policy” on frontier models will lead to a soft realignment of resistance.
Cotopaxi Protest (Google employees, 2017) —> Patagonia Protest (Execs/Boards, 2026).
Etc.
N.B.: Old pal Chaz dropped a a claim about The Office I had never heard. The UK version ran half as long as US one because the former is essentially hopeless and static while the latter reflects American optimism and self-reinvention. While both lack self-awareness, David Brent is surely a darker character than Michael Scarn Scott (especially after Season 1). Americans on average definitely work more. Does that reflect a more upbeat working environment? Just cultural-driven preferences?
A recent Central Air episode was trying to break down the steady productivity growth lead the US has had over Europe for about 20 years. Guest Jason Furman offered the possibility of a higher preference for leisure over income.3 A co-host suggested the rigid labor laws make it hard for people to change jobs and the best companies to make adjustments.4
The latter is consistent with a cultural habituation to pessimism. To resignation to the steady-state.
In contrast to American aspirational immodesty, satisfaction as goal rather than state of being.
Happy 4th.
Prompting WH Chief of Staff to break her social media sabbatical just to rinse off his message.
See, Federal Reserve, on auditioning for Chair of.
I may have missed a nuance here but I thought Ed Prescott had demonstrated this preference is endogenous.
This explanation would at least be consistent with why British Office reflects a fatalist view of career options while Jim can start a sports management company apparently successful enough to pitch Ryan Howard (“eat fresh”).