There’s a race to see what can kill my profession first: artificial intelligence or political stupidity. But I still have to do my job in the meantime.
Let’s start with the AI.
You don’t come to me for insights about AI. I use AI; I have thoughts about AI; I am not an expert in AI. I will say only this: if your takes on AI still involve it hallucinating citations or generating extra fingers in pictures, you’re behind the curve. Benjamin Breen lays out the case for AI (LLMs, specifically chatGPT and Claude) as research tools in history pretty well in this post. His experience tracks with mine: Claude is really good as a tool in many use cases (like making LaTeX tables or drafting abstracts, to take my two least-favorite scholarly tasks). Claude has also at times stunned me with its ability to make recommendations about what to read in the scholarly literature—I mean really good recommendations, on par on with any recommendation I’ve ever gotten from an organic research librarian (sorry, librarians, but it’s true). (Incidentally, this post’s key art is 100% AI-free, or at least I got it from Pexels.)
The hell of it is, of course, that Claude will also absolutely hallucinate.
When I ask it to edit a draft for copy and style, out of ten recommendations, I’d say five are no-brainers to take, one is excellent, two are dumb, and two are blatantly wrong—I mean not even in the original text wrong. (Imagine it editing this piece and telling me to fix the incorrect use of the term “manuscript” in this paragraph.) Is that an enormous advance over not having AI? Eh, no. Is it a pretty good modest advance? Yes. Have I literally felt the LLMs becoming more powerful over the past three years? Undoubtedly.
But the hallucinations are also pretty bad. Once, I asked for something that I needed to help me generate search terms—I needed to know about climatic conditions in the Dakotas in the early 1880s. (Don’t ask why.) This should have been in Claude’s wheelhouse, because the only texts dealing with this should have been really well sorted and verified. (Who fakes information about Bismarck’s climate history?) All I needed was something to help me find more reputable and useful citations—keywords for names of blizzards and that sort of thing.
Claude spun a great yarn. It told me all about the massive shifts in the climate in the 1880s, and even told me about (paraphrasing) the Great Dakota Cold Shift of that period. “Ah, great,” I thought: “keywords!”
Except literally none of that had happened. At all. There was nothing on the Web even remotely close to this (and I do, really, know how to use Google and other databases quite well).
The original had been extremely compelling. If I had been using the machine as a tutor or research assistant, I would have had no reason to doubt it. But it was wrong! Levels of wrong approaching the invention of a new 32nd president of the United States, Percival Carnegie, and his successful mediation of the Indo-Thai war.
A few weeks later, the same thing happened again when I asked for a discussion of the regional politics of the Russian Federation. Again, I needed breadcrumbs to start a search. Again, I asked a tightly polished question and received a brilliant answer about subtle dynamics between the Kremlin and the regions, complete with citations. And again I checked it out and none of it was true. Not a word.
“But didn’t you just say if your AI takes are limited to ‘it hallucinates citations’, you’re out of date?” I hear you ask. Yes, indeed. The trouble with AI right now is that it doesn’t only hallucinate citations—it will also sometimes give you an MA or even (as Breen notes) PhD student level answer. If it only hallucinated, that would be easy; if it only generated PhD-level answers, that would be scary (although in some ways leveling).
I’m not so much of a tech optimist that I believe that all of these problems will be solved soon, or even that some of them can be solved. I can tell you, though, that the experience of using LLMs has gone from novelty to useful tool in less time than it takes to train a human research assistant. And as someone who, myself, has also hallucinated citations, I am aware that human-grade is not a standard of perfection so much as a standard of accountability.
I am, however, enough of a realist to note that the demand for accuracy in education is not so great that an approximately 99% reduction in cost (maybe more!) in learning from and with an LLM compared to a college degree might be appealing to people—even if they end up learning some utterly fabricated concepts. Or, to be more precise, once it seems like expertise is cheaply and plentifully on tap, why would anyone bother to invest in developing human capital? Just use the machine. By analogy: microwave dinners aren’t a perfect substitute for a table at Per Se, but they do a lot more volume in sales.
And, in the meantime, LLMs are making it easier to fake having expertise when it comes to credentialing—or, as we in the biz call it, grading and awarding degrees. Not everyone cheats, but I can assure you that all but the most willfully ignorant instructors are changing what they do in the classroom and in evaluations to account for what feels like a pretty one-sided arms race between honest work and Sam Altman.
Then there’s the politics.
It’s not clear to me that anyone has fully grasped the scale of what is happening. Everything’s paused unexpectedly and even Medicaid isn’t going to be paid one day; the next day, we’re told that almost all of the most egregious holds have been lifted. But it’s less than two weeks in and we already have, arguably, two constitutional crises (birthright citizenship/14th amendment plus whether a president can refuse to spend what Congress authorizes), a grave normative crisis (pardoning the J6 guys), and utterly self-inflicted international crises (e.g. “Green Dawn” and yes this is serious enough for European leaders to meet about contingency plans).
Listing news items, however, isn’t enough to convey the gravity of what’s happening. The OMB pauses on spending, the tariff threats, and the OPM attempts to get rid of large tranches of federal workers are bells that can’t be unrung. They introduce enormous amounts of variance into everyone’s expectations of the future. Moreover, they introduce tremendous uncertainty and friction into the present. Universities (for instance) do a lot of business with the federal government as factories for research and conveyer belts for turning tuition into graduates. If you pause those disbursements, then the science stops and the cash flow dries up. Really fast.
It’s utterly mindblowing to me to hear allegedly serious people talk about DeepSeek’s new AI model as a “Sputnik moment” without following through on how the actual response to actual Sputnik was working seriously and for a generation to develop and extend the federally supported research infrastructure. Instead, right now, the federal government appears bent to break the research pipeline—and that rupture will start with STEM, which depends on federal research dollars much more.
Among the proposals in Project 2025 is an idea to turn the National Institutes of Health into state-funded block grants, which is maybe the most effective way to ruin critical science research funding that I’ve ever heard—and, because of the way university finances operate, at the same time this will also massively wreck the balance sheet of every university that does sponsored (grant-based) research for the feds.
(Why am I so suspicious about this? States do okay, I guess, with things like transportation policy—why not cancer research?
State governments are simply massively less able to attract and retain top talent compared to the federal government because they’re much, much more politicized—well, until now—and pay a lot less. Asking the government of the great state of North Kentucky to run a world-class biomedical research program is prima-facie laughable.
Even “good” state governments, like Massachusetts, are prone to incredible corruption scandals; multiplying the number of granting agencies by 50 probably means multiplying the amount of actual waste, fraud, and abuse by at least twice that factor.
State elected officials, uh, vary widely in their competence. Some are great, yes, but some … Well. Think of the looniest member of the U.S. Congress, and then consider that they represent an altogether more professional breed of politician than the average state legislator. Now ask that legislator to direct a couple hundred million dollars of science research. If it happens—buy shares in producers of ivermectin and laetrile.)
On a more specific level: alliances and international relations prize stability, credibility, and clarity. What a fantastic advert for U.S. hegemony the administration has provided by that standard. It’s one thing to be a mad man in nuclear deterrence; it’s another thing to steer the ship of state into a reef…while it’s docked at a quay. I’m just sure people will be flocking to bandwagon and ally with the country that governs itself as well as this week has gone.
How does this affect me? Well, in the long term, I think that something like universities will turn out okay, even if they may not be the mass institutions of the late 20th century; Oxford, Salamanca, and Bologna are all several hundred years old, and the role of higher education as a socialization tool is something that’s plenty desirable. In the very long term, I’m pretty sure that something like Starfleet Academy will continue to serve our post-scarcity social needs.
In the short term, which is when I do all of my shopping, I’m not sure that we can get through this without a lot of turmoil. Last year, the Biden administration’s Education department screwed up the financial aid application form, and you could feel the nervousness radiating from every university accountant outside of prestige places; would they be forced to eat an enrollment drop or pray that the feds could get their records sorted out in time to pay student assistance? Now, imagine that the administration decides to just … pause federal spending on science or student loans or whatever until universities certify they’re 100% woke-free. Sure, there’s going to be a judge who steps in to remedy this … but in the meantime that cash flow has frozen up while expenses keep going, and suddenly the institution is either floating some short-term loans or digging into its seed corn.
And that’s before we get into the major, potentially legislature-approved changes to the sector, like taxing all the big endowments to pay for an online university. (Sounds good from some directions, but let’s remember who’d be implementing this.)
Also in the short term, I very much do not live in the United States and I suddenly have a very keen interest in the U.S. not going full isolationist—or manifesting its Greenlandic destiny—while I’m living out here.
And in the even shorter term, gah, do you know how many news articles I have to read to keep up with my twice-weekly Congress class? “Oh, just don’t read the news,” say people fortunate enough to not have a job that is, in times like these, functionally The News: Explained. Yeah, okay, I won’t lie, it’s invigorating and definitely helps make the case for the relevance of course materials, but it’s also so darn much all the time.
In the meantime, there’s a real air of unreality. I go to my little office, type my little papers, and teach my little lessons. I offer bons mots about the burning of the Constitution and explain What It All Means for the international system. I ask Claude to check if I made any subject-verb errors. It’s like LARPing in a giant game of Dark Academia! It’s fun, it’s cozy, and it helps me ignore the screams and thuds I can hear in the distance.
“Who fakes information about Bismarck’s climate history?” LOL! The column reads like stand-up for, shall we say, high-information voters. Will you be here all week?
Paul, why where you researching the dakotas in the 1880s