Do (human) readers prefer AI writers?
It seems so, do read through the whole abstract:
The use of copyrighted books for training AI models has led to numerous lawsuits from authors concerned about AI’s ability to generate derivative content. Yet it’s unclear whether these models can generate high quality literary text while emulating authors’ styles/voices. To answer this we conducted a preregistered study comparing MFA-trained expert writers with three frontier AI models: ChatGPT, Claude, and Gemini in writing up to 450 word excerpts emulating 50 awardwinning authors’ (including Nobel laureates, Booker Prize winners, and young emerging National Book Award finalists) diverse styles. In blind pairwise evaluations by 159 representative expert (MFA-trained writers from top U.S. writing programs) and lay readers (recruited via Prolific), AI-generated text from in-context prompting was strongly disfavored by experts for both stylistic fidelity (odds ratio [OR]=0.16, p < 10^-8) and writing quality (OR=0.13, p< 10^-7) but showed mixed results with lay readers. However, fine-tuning ChatGPT on individual author’s complete works completely reversed these findings: experts now favored AI-generated text for stylistic fidelity (OR=8.16, p < 10^-13) and writing quality (OR=1.87, p=0.010), with lay readers showing similar shifts. These effects are robust under cluster-robust inference and generalize across authors and styles in author-level heterogeneity analyses. The fine-tuned outputs were rarely flagged as AI-generated (3% rate versus 97% for incontext prompting) by state-of-the-art AI detectors. Mediation analysis reveals this reversal occurs because fine-tuning eliminates detectable AI stylistic quirks (e.g., cliché density) that penalize incontext outputs, altering the relationship between AI detectability and reader preference. While we do not account for additional costs of human effort required to transform raw AI output into cohesive, publishable novel length prose, the median fine-tuning and inference cost of $81 per author represents a dramatic 99.7% reduction compared to typical professional writer compensation. Author-specific fine-tuning thus enables non-verbatim AI writing that readers prefer to expert human writing, thereby providing empirical evidence directly relevant to copyright’s fourth fair-use factor, the “effect upon the potential market or value” of the source works.
That is from a new paper by Tuhin Chakrabarty, Jane C. Ginsburg, and Paramveer Dhillon. For the pointer I thank the excellent Kevin Lewis. I recall an earlier piece showing that LLMs also prefer LLM outputs?
John Cochrane understands the elasticity of supply
Everyone is focused on building, but “supply” is so much more than building. There is tremendous supply in using more efficiently what we have now. Most cities have laws against renting parts of single family homes, or sharing larger homes. Think how many spare bedrooms are empty every night. There is plenty of housing supply in the US, it’s just not in places where people want to move. Others moving out is “supply,” and greatly impeded. Older people stay in too-big houses and apartments, in locations close to work and school opportunities that young families desire, but the older people no longer need. Why? If they sell, they are taxed on capital gains, even just due to inflation. They lose property tax exemptions, and, of course, rent control protection. Each older person who cashes in, downsizes, or moves to a more neighborhood more suited to them, supplies a house or apartment. The non-portable fixed rate 30 year mortgage, an invention of our federal housing subsidy regime, leads people to stay where they are rather than move to where they want to go, and free up a scarce house or condo for someone else. Strong apparently “consumer protection” laws in rental contracts dry up the supply, especially to the marginalized. If you can’t kick people out, you’re much more careful who you let in. Limits on short term rentals limit rentals. Remove rent controls, permanently, and houses and condos can be rented. Many houses and apartments need rehab, not new construction, which can happen very quickly once owners know they will not be robbed of their investment. Even “affordable” housing leads people to stay where they are, rather than move to better opportunities for them and free up an apartment for someone else, because it’s rationed with long waiting lists.
The rest of the post is an excellent analysis of rent control. On one of John’s closing points, I should note I am not opposed to speculating about motivations.
Monday assorted links
1. Andrej: “Software 1.0 easily automates what you can specify. Software 2.0 easily automates what you can verify.”
2. Transgenerational Epigenetic Inheritance.
3. “The national [cow] herd is now the smallest since the 1950s.” (FT)
4. California housing remains stalled.
Illegal Immigrants Didn’t Break the Housing Market; Bad Policy Did
In an interview, JD Vance claimed:
[H]ousing is way too expensive….because we flooded the country with 30 million illegal immigrants who were taking houses that ought by right go to American citizens.
I noted on Twitter that this framing reeks of socialist thinking, national socialist to be precise. A demand for the state to designate a privileged class that get special rights to scarce goods. Treating housing as a fixed stock to be allocated to a favored in-group while blaming an out-group for shortages is collectivist politics driven by grievance, not market reasoning. In short, grievance and entitlement, zero-sum thinking and central planning wrapped into one ugly bundle.
That criticism set people off. The first rebuttal was predictable “Ha ha, the economist forgot about supply and demand!”—a miss, because my point wasn’t about the mechanics of house-price growth but about Vance’s rhetoric: the collectivism and the cheap politics of blaming outsiders. The second rebuttal was that “America belongs to Americans” so of course illegal immigrants shouldn’t be allowed to buy homes.
The second objection is amusing because who is harmed most when a government bans immigrants from buying homes or deports a chunk of potential buyers? American home sellers. The way such bans “work” is by preventing sellers from accepting the highest bid. In effect, these policies are a tax on sellers combined with a subsidy to a subset of buyers.
So bans on foreign buyers are really about taxing some Americans and subsidizing others. Moreover, although the economic logic of illegals pushing up demand is sound, the numbers don’t add up to much. First, there aren’t 30 million illegals; the best estimates are roughly 14 million. And second illegals are obviously not the reason homes blow past a million dollars in places like San Francisco, San Jose, Washington, or New York! The effect of illegal immigrant on house prices exists but is small—the bigger factors are native population growth, rising incomes, zoning rules, and strict limits on new construction. Block illegal immigrants from buying homes and you will get a pause in price growth, but once demand from natives keeps rising against a capped supply, prices will climb back to where they were.
That gets to the deeper problem with Vance’s style of thinking. If “fixing” housing scarcity means blaming whichever group is politically convenient, you end up cycling through targets: illegal immigrants first, then legal immigrants (as Canada has done), then the children of immigrants, then wealthy buyers, then racial or religious minorities. Indeed, one wonders if the blame is the goal.
If you actually want to solve the problem of housing scarcity, stop the scapegoating and start supporting the disliked people who are actually working to reduce scarcity: the developers. Loosen zoning and cut the rules that choke what can be built. Redirect political energy away from trying to demolish imagined enemies and instead build, baby, build.
Wise Words Addendum (hat tip G. Scott Shand):
There is a cultural movement in the white working class to blame problems on society or the government, and that movement gains adherents by the day….We’ll get fired for tardiness, or for stealing merchandise and selling it on eBay, or for having a customer complain about the smell of alcohol on our breath, or for taking five thirty-minute restroom breaks per shift. We talk about the value of hard work but tell ourselves that the reason we’re not working is some perceived unfairness: Obama shut down the coal mines, or all the jobs went to the Chinese. These are the lies we tell ourselves to solve the cognitive dissonance—the broken connection between the world we see and the values we preach.
There is no great stagnation (not any more — really!)
A remote-controlled robot the size of a grain of sand can swim through blood vessels to deliver drugs before dissolving into the body. The technology could allow doctors to administer small amounts of drugs to specific sites, avoiding the toxic side effects of body-wide therapies.
…The system has yet to be trialled in people, but it shows promise because it works in a roughly human-sized body, and because all its components have already been shown to be biocompatible, says Bradley Nelson, a mechanical engineer at Swiss Federal Institute of Technology (ETH) in Zurich, who co-led the work.
We will see, but it is wonderful that such an idea is even in the running. Here is the full article, via A.J.
“May I meet you?”
Bill Ackman suggests that opener as a way for men to meet women, and notes it worked for him when he was younger and unmarried. Like this: “I would ask: “May I meet you?” before engaging further in a conversation. I almost never got a No. It inevitably enabled the opportunity for a further conversation. I met a lot of really interesting people this way. I think the combination of proper grammar and politeness was the key to its effectiveness. You might give it a try.”
In response, a bunch of people have shrieked that he is a billionaire (he was not then, though perhaps he had Aristotelian billionaire potentiality?), that he is six foot three (he probably was tall back then too), and that he is good looking. Or perhaps effective meeting and dating strategies have changed?
I readily admit I am well below average in this and all related areas concerning either meeting strangers or chatting up women, whether it concerns knowledge or praxis. But I have an opinion nonetheless.
I observe that so many young men these days just do not make much effort at all. They do not approach women with any sort of opening line, whether in person or through apps. If this gets them off the zero point, it is almost certainly a good thing. Maybe it is bad tactics for some people, if only because you are too nerdy and cannot deliver the words with the right charming tone. So be it. The young men with that problem can then adjust and try it some other way. It is still a plus to get them thinking about opening lines at all, and to think about meeting women at all. So I am fully on board with Bill’s suggestion. He never said that is all you should be doing, or to make that your main thing. It is unlikely that his suggestion is the best thing you could be doing, think of it simply as pressing the “activation button” on seeking a partner.
It is a bit like my advice on writing. Your big enemy is not “I did not get enough written today.” Rather it is “I did not write today at all.” That point applies to so many different aspects of life. Discrete choice econometrics!
Addendum: Bill adds that it works better when you are moving. Let’s avoid this equilibrium. And here are some other comments, I am not sure of the proper attribution.
Sunday assorted links
1. Julia Steinberg on Solano County.
2. Mentorship advice from Avital Balwit.
3. Orcas are attacking boats in coordinated fashion. And more.
4. James Brown, Night Train, live.
6. Why don’t people return their shopping carts? (Often I am not sure it is my obligation to? In that context non-return might be Coasean.)
7. From Eliezer.
Why are US Clinical Trials so Expensive?
Dave Ricks, CEO of Eli Lilly, speaking on the excellent Cheeky Pint Podcast (hosted by John Collison, sometimes joined by Patrick as in this episode) had the clearest discussion of why US clinical trial costs are so expensive that I have read.
One point is obvious once you hear it: Sponsors must provide high-end care to trial participants–thus because U.S. health care is expensive, US clinical trials are expensive. Clinical trial costs are lower in other countries because health care costs are lower in other countries but a surprising consequence is that it’s also easier to recruit patients in other countries because sponsors can offer them care that’s clearly better than what they normally receive. In the US, baseline care is already so good, at least at major hospital centers where you want to run clinical trials, that it’s more difficult to recruit patients. Add in IRB friction and other recruitment problems, and U.S. trial costs climb fast.
Patrick
I looked at the numbers. So, apparently the median clinical trial enrollee now costs $40,000. The median US wage is $60,000, so we’re talking two thirds. Why and why couldn’t it be a 10th or a hundredth of what it is?David (00:10:50):
Yeah, brilliant question and one we’ve spent a lot of time working on…“Why does a trial cost so much?” Well, we’re taking the sickest slice of the healthcare system that are costing the most. And we’re ingesting them. We’re taking them out of the healthcare system and putting them in a clinical trial. Typically we pay for all care. So we are literally running the healthcare system for those individuals and that is in some ways for control, because you want to have the best standard of care so your experiment is properly conducted and it’s not just left to the whims of hundreds of individual doctors and people in Ireland versus the US getting different background therapies. So you standardize that, that costs money because sort of leveling up a lot of things, but then also in some ways you’re paying a premium to both get the treating physicians and have great care to get the patient. We don’t offer them remuneration, but they get great care and inducement to be in the study because you’re subjecting yourself quite often, not all the case, but to something other than the standard of care, either placebo or this. Or, in more specialized care, often it’s standard care plus X where X could actually be doing harm, not good. So people have to go into that in a blinded way and I guess the consideration is you’ll get the best care.Patrick (00:12:51):
Of the $40,000. How much of that should I look at as inducement and encouragement for the patient and how much should I look at it as the cost of doing things given the regulatory apparatus that exists?David (00:13:02):
The patient part is the level up part and I would say 20, 30% of the cost of studies typically would be this. So you’re buying the best standard of care, you’re not getting something less. That’s medicine costs, you’re getting more testing, you’re getting more visits, and then there is a premium that goes to institutions, not usually to the physician, the institution to pay for the time of everybody involved in it plus something. We read a lot about it in the NIH cuts, the 60% Harvard markup or whatever. There’s something like that in all clinical trials too. Overhead coverage, whatnot. But it’s paying for things that aren’t in the trial.Patrick (00:13:40):
US healthcare is famously the most expensive in the world. Yes. Do you run trials outside the US?David (00:13:44):
Yeah, actually most. I mean we want to actually do more in the US. This is a problem I think for our country. Take cancer care where you think, okay, what’s the one thing the US system’s really good at? If I had cancer, I’d come to the US, that’s definitely true. But only 4% of patients who have cancer in the US are in clinical trials. Whereas in Spain and Australia it’s over 25%.And some of that is because they’ve optimized the system so it’s easier to run and then enroll, which I’d like to get to, people in the trials. But some of it is also that the background of care isn’t as good. So that level up inducement is better for the patient and the physician. Here, the standard’s pretty good, so people are like, “Do I want to do something where there’s extra visits and travel time?” There’s another problem in the US which is, we have really good standards of care but also quite different performing systems and we often want to place our trials in the best performing systems that are famous, like MD Anderson or the Brigham. And those are the most congested with trials and therefore they’re the slowest and most expensive. So there’s a bit of a competition for place that goes on as well.
But overall, I would say in our diabetes and cardiovascular trials, many, many more patients are in our trials outside the US than in and that really shouldn’t be other than cost of the system. And to some degree the tuning of the system, like I mentioned with Spain and Australia toward doing more clinical trials. For instance, here in the US, everywhere you get ethics clearance, we call it IRB. The US has a decentralized system, so you have to go to every system you’re doing a study in. Some countries like Australia have a single system, so you just have one stop and then the whole country is available to recruit those types of things.
Patrick (00:15:31):
You said you want to talk about enrollment?David (00:15:32):
Yeah, yeah. It’s fascinating. So drug development time in the industry is about 10 years in the clinic, a little less right now. We’re running a little less than seven at Lilly, so that’s the optimization I spoke about. But actually, half of that seven is we have a protocol open, that means it’s an experiment we want to run. We have sites trained, they’re waiting for patients to walk in their door and to propose, “Would you like to be in the study?” But we don’t have enough people in the study. So you’re in the serial process, diffuse serial process, waiting for people to show up. You think, “Wow, that seems like we could do better than that. If Taylor Swift can sell at a concert in a few seconds, why can’t I fill an Alzheimer’s study? There seem to be lots of patients.” But that’s healthcare. It’s very tough. We’ve done some interesting things recently to work around that. One thing that’s an idea that partially works now is culling existing databases and contacting patients.Patrick (00:16:27):
Proactive outreach.
See also Chertman and Teslo at IFP who have a lot of excellent material on clinical trial abundance.
Lots of other interesting material in this episode including how Eli Lilly Direct—driven largely by Zepbound—has quickly become a huge pharmacy. The direct-to-consumer model it represents could be highly productive as more drugs for preventing disease are developed. I am not as anti-PBM as Ricks and almost everyone in the industry are but I will leave that for another day.
Here is the Cheeky Pint Podcast main page.
Best fiction of 2025
Solvej Balle, On the Calculation of Volume I, Volume II. Volume III is due out in English late this year I have read it already in German. A very strong series, reading ahead in German is a good demonstration of how much I like them.
Suat Dervis, The Prisoner of Ankara. A Turkish novel from mid-century, in English for the first time.
Emmanuel Carrere, V13: Chronicle of a Trial. Non-fiction but it is more likely reading fiction, it just happens to be true (supposedly).
Alain Mabanckou, Dealing with the Dead. Most African fiction does not connect with me, and there is a tendency for the reviews to be untrustworthy. This “cemetery memoir,” from the Congo (via UCLA), held my interest throughout.
Kiran Desai, The Loneliness of Sonia and Sunny.
Eça de Queiros, Adam and Eve in Paradise. Originally from the 19th century, but translated into English only this year. A 60 pp. novella about exactly what the title indicates, noting that matters are not as simple as the first telling of that story might have suggested.
The Poems of Seamus Heaney. Not yet received, but obviously this is a winner.
Overall, the Balle, Desai, and Heaney make for very strong entries, so this was a good year for fiction.
Emergent Ventures winners 48th cohort
Minh Nguyen, Dallas, 16, to address problems of space debris.
Serena Chen and Valerie Wu, Taipei and Los Angeles, better blood transportation.
Griffin Li, Boston, 19, general career support, issues in market pricing on apps.
Jamie Croucher, London, for a new scientific society, Society for Technological Advancement.
Sri Anumakonda, Carnegie Mellon, robotic hands.
Ben James, London, lowering the price of electricity.
Lucio Aurelio Moreira Menezes, Fortaleza and Stanford, computational biology.
Lauren Gilbert, London, a progress-oriented magazine for the developing world?
Annika Baberwal, Dubai, 8th grade, helping race cars run faster.
Bernardo Melotti, Bologna/Stanford, equipment to speed biological advances.
Elyas Sepahi, London, mathematics and machine learning.
Harishvin Sasikumar, Vancouver, 17, brain interfaces.
The Effect of Video Watching on Children’s Skills
This paper documents video consumption among school-aged children in the U.S. and explores its impact on human capital development. Video watching is common across all segments of society, yet surprisingly little is known about its developmental consequences. With a bunching identification strategy, we find that an additional hour of daily video consumption has a negative impact on children’s noncognitive skills, with harmful effects on both internalizing behaviors (e.g., depression) and externalizing behaviors (e.g., social difficulties). We find a positive effect on math skills, though the effect on an aggregate measure of cognitive skills is smaller and not statistically significant. These findings are robust and largely stable across most demographics and different ways of measuring skills and video watching. We find evidence that for Hispanic children, video watching has positive effects on both cognitive and noncognitive skills—potentially reflecting its role in supporting cultural assimilation. Interestingly, the marginal effects of video watching remain relatively stable regardless of how much time children spend on the activity, with similar incremental impacts observed among those who watch very little and those who watch for many hours.
That is from a new NBER working paper by
Saturday assorted links
1. John Brennan clarifies on UAPs.
2. “For the rest of the world, including Europe, wide adoption of US dollar stablecoins for payment purposes would be equivalent to the privatization of seigniorage by global actors,” notes Hélène Rey, an economist, who fears rising “tax evasion” and “lower demand for non-US government bonds”. FT link here.
3. “Puffins have been seen on the Isle of Muck in County Antrim for the first time in years, after a major scheme to remove invasive brown rats.” Link here.
5. Is there some economic recovery in Nigeria?
6. Charles Murray on finding religion.
7. Ross Douthat with Paul Kingsnorth (NYT).
The spatial construction of Incan monies?
Money can be many things, but this to me is a new one. Or is it?
Stretching for 1.5km and consisting of approximately 5200 precisely aligned holes, Monte Sierpe in southern Peru is a remarkable construction that likely dates to at least the Late Intermediate Period (AD 1000–1400) and saw continued use by the Inca (AD 1400–1532). Yet its function remains uncertain. Here, the authors report on new analyses of drone imagery and sediment samples that reveal numerical patterns in layout, potential parallels with Inca knotted-string records and the presence of crops and wild plants. All this, the authors argue, suggests that Monte Sierpe functioned as a local, Indigenous system of accounting and exchange.
That is newly published research from Jacob L. Bongers, et.al. Via the excellent Kevin Lewis.
Was Brexit worse than we had thought?
This paper examines the impact of the UK’s decision to leave the European Union (Brexit) in 2016. Using almost a decade of data since the referendum, we combine simulations based on macro data with estimates derived from micro data collected through our Decision Maker Panel survey. These estimates suggest that by 2025, Brexit had reduced UK GDP by 6% to 8%, with the impact accumulating gradually over time. We estimate that investment was reduced by between 12% and 18%, employment by 3% to 4% and productivity by 3% to 4%. These large negative impacts reflect a combination of elevated uncertainty, reduced demand, diverted management time, and increased misallocation of resources from a protracted Brexit process. Comparing these with contemporary forecasts – providing a rare macro example to complement the burgeoning micro-literature of social science predictions – shows that these forecasts were accurate over a 5-year horizon, but they underestimated the impact over a decade.
That is from a new NBER working paper by
*The Science of Second Chances*
The author is economist Jennifer Doleac, and the subtitle is A Revolution in Criminal Justice. Excerpt:
We found that adding anyone charged with a felony to the law enforcement DNA database in Denmark reduced future criminal convictions by over 40 percent. Again, people responded to the higher probability of getting caught by committing fewer crimes. Being added to the database also increased enrollment in school and rates of employment — signs that folks really were on a better path. This effect was largest for the youngest men, those ages eighteen to twenty-four.
Incentives matter. An excellent book, recommended, due out next year.