You’ve probably seen that Canadian PM Mark Carney has walked away from trade talks with the US. That has led the US to impose staggeringly large (and certainly illegal) tariffs on Canada and Canada is responding in kind. It’s not clear precisely what demands the US made that prompted Carney to do this. Given the pain this will cause Canada, at least in the near-term far more than the US, they must have been not only unacceptable but an affront to Canada’s sovereignty and national dignity.
As I understand it, negotiations appeared to be going well and productively. They began as an attempt to find an agreement to forestall Trump’s original threat of a new round of draconian and illegal tariffs. Indeed, they appeared to be so near completion that an article I saw mid-week referred to Mexico eyeing and looking to negotiate a deal with comparable contours. Point being, it was a measure not only of how well the negotiations appeared to be going but how close they seemed to completion. The broad outlines were known to other interested countries.
Then Carney abruptly ended them in response to new demands.
I wanted to follow up on my post about Pat Ruffini’s deep dive about the accuracy of Senate race polling in recent cycles and whether they’ve overstated Democrats’ strength. But since I wrote that post, I’ve also gotten a few emails from friends and TPM readers about polling generally. So I wanted to combine all of this together into some general comments on polling and whether we should care about it at all.
In the wake of the big apparent polling misses in Michigan and Wisconsin, we’re seeing another wave of discussion about a purported “crisis” in polling and what we should collectively do about it or what the polling industry should do about it. Mostly, the answer is there is no “crisis” — at least in the sense of one that any of us should care about as some problem for American democracy or elections or anything else. If you’re in the polling business, it may well be a problem. But if you’re not, who cares? There might be a crisis in Swiss cheese marketing too. But if I’m not a cheese monger, why do I care?
If you’ve been online in the last few days you’ve undoubtedly been exposed to Natalie Harp Discourse (NHD), the phenomenon surrounding a woman who has served since January 2025 as President Trump’s human emotional comfort blanket. Attention to Harp had been growing online over the last couple weeks. (Perhaps the canonical account on her role was published by the New York Times just after the 2024 election.) Even the TPM Editors’ Blog has been in on the action. But it kicked into high gear on Sunday when Sen. Jon Ossoff (D-GA) referenced her with a mere three words in a campaign speech which was then sliced and diced for maximum social media exposure. Let me quote the full sentence with the three words about Harp in italics: “He golfs and trades stocks. See, he doesn’t want to do the job. He wants to build his ballroom and travel with Natalie on their apparently defenseless flying palace gifted by the emir of Qatar.”
This may seem like fluff and frivolity to some. But among other things it brings to the surface a question that has hovered around the opposition to Trump ever since he entered the political scene more than a decade ago. I had been thinking about this in recent weeks because it becomes increasingly relevant, especially with the onset of the 2026 midterms.
In my last post I mentioned this data analysis GOP pollster and consultant Pat Ruffini has put together looking at Senate polling going back through the last four cycles, so 2018 through 2024. Ruffini is a very partisan Republican in all the meanings of that word. But I think of him as broadly reliable when it comes to compiling and analyzing hard data. His analysis, probably not surprisingly, shows that polling in aggregate over-estimated Democrats in each cycle. Needless to say, there’s a lot of “cope” in this analysis, reassuring fretful Republicans that it’s not as bad as it looks. But I was curious because this at least broadly matches with my observations. Not always but more often than not.
You can read the analysis here. You have to subscribe to his Substack to read the whole thing. But you can get the gist before you get to the paywall. I’ve reached out to several polling analysts whose judgment I trust to get their sense of the analysis. So perhaps I’ll have more to say based on responses.
Today I’m going to discuss two tidbits from the world of polling. In a second post I’m going to discuss an interesting and, for Democrats, somewhat chastening look at the last four cycles of Senate polling compiled by GOP pollster/consultant Pat Ruffini. But first some stunning news today about an out-of-nowhere polling outfit called Median Strategies. They polled a number of races but their survey showing DSA-backed Wisconsin gubernatorial candidate Francesca Hong with a 23-point lead got the most attention. Today the firm announced it is ceasing operations, says it is withdrawing all its previous polls and says they should not be treated as “genuine polling data.”
They then go on to make this almost unbelievable statement: “Median Strategies was created as a short-term social experiment examining how purported polling information could enter and spread through the political information ecosystem wihtout independent verification.” They then follow with a “Thank you for your understanding.”
Here is an update on a story you may not yet have heard of. It manages to be at once fascinating, deeply disturbing and a perfect living metaphor of our age. AI models are vulnerable to something called “model collapse.” In the early days of LLM-based AI there was a seemingly limitless store of online content to train it on. But that’s mostly been used up. Meanwhile, a rapidly growing percentage of online content is produced by AI. If you try to train AI on the product of AI it’s a bit like locking a person in a sealed room and watching them slowly suffocate as they breath in the product of their cellular respiration (CO2) rather than the fuel (oxygen) that makes the body function. Not a perfect analogy but you get the idea. This has made the creators of AI increasingly desperate in their search for new artifacts of human intellection — a phrase I heard once from the great Chaucer scholar John V. Fleming who died this past May — to feed into large language model AI systems.
The solution? Physical books published before 2022. You can guarantee as a matter of absolute certainty that whatever their quality, their merits, whatever they’re about, they are the product of human minds.
It’s relatively easy to scan a book without damaging it. But if you’re doing it at scale the costs of scanning and digitizing a book’s contents without damaging it are vastly higher. So what AI companies appear to be doing is buying up books, cutting off the spines, feeding them into industrial scale scanners and then throwing away.
