We are one month away from the midterm elections and I wanted to put my marker down on what I think is going to happen there. The good news for Democrats is that I think that they will take control of the House and, narrowly, the Senate. When the dust has settled I expect Democrats to hold about between 218 and 230 seats in the House and 51 seats in the Senate.
The bad news for Democrats is that it will not be a “wave election” and their majorities will be small and fractious. I define a “wave” as 20 or more House seats.
As evidence for that view I would suggest reading Bryan Bennett’s post at Ruy Teixeira and Michael Bahareen’s Substack. The short version is that the Democratic “brand advantage” is sufficiently low that they are unlikely to overperform it but rather to underperform it. From the linked Substack:
The x-axis is “relative brand advantage” (RBA), which calculates how much higher or lower favorability for the Democratic Party is compared to the Republican Party. For example, in 2006, the average net favorability of the Democratic Party was net +13 and the average net favorability of the Republican Party was net -9; thus, the RBA for Democrats was +22. In 2018, the average net favorability of the Democratic Party was net +1 and the average net favorability of the Republican Party was net -16; thus, the RBA for Democrats was +17. In Q3 of 2026, the average net favorability of the Democratic Party is net -13 and the average net favorability of the Republican Party is net -17; thus, the relative brand advantage for Democrats is just +4. This post does not dwell on the deterioration of the Democratic Party brand, of which I’ve also written about extensively and expressed many concerns about, but I will say this: despite having a narrow RBA, favorability of the Democratic Party dropping by a net 26 points over the last two decades is deeply problematic and should not be overlooked.
and
When charted this way, Democrats have tended to underperform or equal their RBA in the generic ballot—and, ultimately, in the actual election results.
In others words an extraordinary presidency may produce an ordinary midterm correction.
The greater question is how will the Democrats use their new majorities? Specifically, will they devote their efforts to “fighting Trump”, will they focus on tiny minorities, or will they do their best to help ordinary Americans? I’m hoping it will be the last but I’m concerned the Democratic leadership will not be able to subdue or even resist their most extreme subfactions.
Here’s a modest proposal for giving LLM AI service providers exactly what they want:
Delay settling the rules until management and early investors have had an opportunity to issue an IPO and cash out.
Add a new agency or committee to provide ample opportunity for capture.
Grandfather existing providers and models. Apply the expensive new requirements principally to future entrants. Being first then becomes a government-protected advantage.
Make compliance expensive regardless of company size. Require licensing, specialized counsel, extensive documentation, audits, and recurring certification. A fixed compliance bill is manageable for a giant and prohibitive for a prospective competitor.
Let the leading providers define “safe AI.” Turn their existing practices, evaluation methods, and organizational structures into mandatory standards. Competitors must reproduce the incumbents’ overhead before offering an alternative.
Make regulatory compliance a shield against liability. Once a provider has completed the prescribed paperwork, restrict injured parties’ ability to recover damages. Certification becomes protection for the company rather than protection for the public.
Place responsibility on customers and downstream developers. Give providers broad latitude to disclaim responsibility through their terms of service, even where customers have little ability to inspect or control the underlying system.
Regulate models rather than harmful conduct. Require permission to develop or release a general-purpose tool, instead of concentrating enforcement on fraud, negligence, discrimination, and other identifiable harms. That makes the regulator a gatekeeper to the market.
Make open models and local operation presumptively suspect. Require centralized monitoring, identity verification, remote shutdown, and continuing provider control. Make running a model yourself legally difficult enough that renting access from an approved provider becomes the practical default.
Preempt state remedies before establishing an effective federal substitute. Remove alternative avenues of enforcement while leaving the national framework unfinished, weak, or dependent on industry cooperation.
Additional points gratefully accepted. Extra points for providing an advantage to foreign competitors.
The editors of the Washington Post provide a list of alternatives for reducing the federal deficit and a handy questionnaire for evaluating your own preferences. They write:
They characterized my responses as “Pragmatic Balancer”:
You cut spending and you raised taxes, spreading the adjustment widely enough that no single group carries it alone. It’s an approach designed to be politically possible while getting the situation under control.
I think they omitted a number of key points. First, there is empirical evidence that a large public debt overhang impedes economic growth. That rules out the easy approach: just outgrow the debt. We cannot safely assume growth will rescue us, especially when accumulated debt may itself weaken growth.
The second is that the graph they display overstates how frequently we have balanced the budget. It conceals how dependent those results were on a temporary demographic advantage. In only one year did the Clinton Administration produce a substantial budget surplus without counting Social Security’s surplus. That should give us pause. The opposite is the case now.
I certainly hope that Donna Brazile is right in her evaluation of the impact of prospective Democratic Socialist members of the Democratic caucus in the next Congress as she wrote in her op-ed in the Washington Post:
Contrary to ridiculous fearmongering by President Donald Trump, House Minority Leader Hakeem Jeffries (D-New York) will not let democratic socialists turn America into a communist nation if he becomes speaker. And he won’t let them control the House, either.
Here are the DSA policies she points to in the op-ed:
The DSA has drawn outsize media attention because its platform is so extreme. It calls for “public ownership of the largest corporations and essential industries,” defunding the Defense Department, “fully abolishing the police and prison system,” abolishing the Senate, “Housing For All,” “universal healthcare at no cost to individuals,” “a federal jobs guarantee” and “amnesty for all immigrants regardless of status,” among other demands. Trump and Republicans are trying to convince voters that most Democrats agree with these radical views. Democrats must show that they don’t. Jeffries has already said that he doesn’t support the DSA agenda.
and concludes:
Once serving in Congress, democratic socialists will realize that advocating for pie-in-the-sky proposals accomplishes nothing. They will instead need to work with their fellow Democrats to make less radical change.
I hope she’s right. I’m concerned that she may be surprised at how the party has changed since she was Chair of the Democratic National Committee. At least one of the measures she points to, “Medicare for All”, has the support of the majority of Democratic House members. Others of those measures or incremental steps in their direction may have more support than she recognizes.
As we have learned in recent years having a relatively small number of members does not mean that their proposals won’t hold an outsize influence on the caucus as a whole. The threat of primary challenges does wonders to focus the mind.
After cruising the Inland Passage for several days our ship docked at Whittier and we disembarked for good. We tipped our cabin steward well for taking such good care of us (and under the assumption that much of what he earned was from tips).
Then began what was the most memorable part of our trip—staying with friends in Wasilla, outside Anchorage. I don’t believe I’ve mentioned that we were traveling with a party of eight. We were traveling with our dog breeders and others my wife had made friends with attending Jack’s shows. One of the great things about “the fancy” (that’s what those who are serious about dogs is called in aggregate) is that you encounter and make friends with people from all over the country. Among those were people in Wasilla.
Over the next several days I will post on our experiences there.
Wild video shows a cow being airlifted to safety after getting stranded on a French mountain.
Vahine the bovine got stuck in the mud while grazing in the Pyrenees on Sept. 11 and couldn’t make her way down the steep terrain, which also couldn’t be accessed by car.
I can’t link to the video or otherwise I’d embed it—it’s embedded in the story.
It’s quite gratifying to see the argument I’ve been making for years gradually start seeping into common recognition. In this case it’s that LLM AI service providers (and their stockholders) should bear strict liability. Gabriel Rauterberg and Sarath Sanga remark in an op-ed at the Wall Street Journal:
The heads of frontier labs have said their companies could be responsible for such a mass tort, or worse. They’ve made the case against limited liability.
That case, as we argue in a recent paper, is strongest when a corporation’s profits and the risks it poses rise together, and when those risks are extreme and unbounded. That’s exactly the situation AI executives describe. Customers demand the capabilities that make the product so dangerous. The more successful the product, the more risk likely lands on the public.
Building more-capable models without these risks is the engineering problem that labs say they need more time to solve. Congress isn’t about to solve it. But Congress doesn’t have to. It can establish a liability rule and let incentives do the work.
In the 19th century, locomotives threw off hot embers that set nearby fields and barns ablaze. Lawmakers didn’t engineer a technical solution. They an
That should be easy for Congress to achieve even in a divided Congress.
Building better models is only one of the engineering problems AI service providers face. They must also find ways to deploy them safely. Strict liability for harms their services cause would give them a powerful incentive to do so. And if the potential harm exceeds what the company can pay, limiting the loss to the company’s assets leaves the public bearing the risk while shareholders retain the upside. That is the part of the authors’ argument Congress should take seriously and the sooner the better.
It used no moving or still images of Garbo: it was created solely by providing prompts to AI models.
Hence the title of this post. Why “cast” a completely fictional AI-generated actress when you can “cast” an actress who was once alive and has a fanbase?
I’ve been predicting this since 2004 when I saw Sky Captain and the World of Tomorrow which used the image of the deceased Laurence Olivier (with the permission of his estate). I expected it to be created by fans, recasting classic movies with other performers that had been suggested, e.g. Ronald Reagan instead of Humphrey Bogart as Rick in Casablanca. I certainly didn’t expect to see Greta Garbo hawking ball bearings. I suppose I shouldn’t be surprised.
Engaging with the printed word in book form, Marriott writes, “changed how people thought.” Long-form prose strengthened the mental muscles necessary for prolonged concentration on arguments and complexities: Readers, Marriott says, began to question received ideas: “Where does power come from? How should we classify whales? How do we know the Bible is true?”
Today, numerous literature professors report that they formerly expected students to read a novel a week, but no longer assign such masterpieces as George Eliot’s “Middlemarch” or James Joyce’s “Ulysses.” Students are not just unwilling, they are unable to comprehend sophisticated, nuanced narratives. As deep reading becomes, Marriott says, a “rare, even esoteric accomplishment,” our culture becomes unmoored “from its foundations in the written word.”
I wish I could say “Welcome to the fight” but I’m afraid that “the fight” is already over and we lost. You can see it all around you in our political discourse: the agonistic modes of expression, absence of logical arguments, division into tribes.
He’s actually understating the matter. It’s more than “strengthening a muscle”. There are actual, physical differences between the brains of the literate and the post-literate, as I have documented in my many posts over the last 20 years on what I’ve deemed “visualcy”.
Learning to read changes the brain, and sustained reading gives us repeated practice in following arguments that unfold beyond the length of a clip or a slogan. We know much less about what happens when a literate society stops making that practice commonplace. The question is no longer simply whether people can read the words. It is whether they can derive meaning from an extended written argument.
We left Skagway and spent the next day or so at sea, sailing north through the Inland Passage. We viewed the fjords and glaciers, watching for ocean wildlife.
