Category: Uncategorized
Those new service sector jobs?
An AI memory startup called Memvid is offering $800 for a one-day, eight-hour shift for one candidate to “bully” AI chatbots by telling them what to do on camera.
Business Insider reported this week that Memvid wants someone to spend eight hours testing and critiquing the memory of popular AI chatbots, effectively paying $100 an hour for what they have branded as a “professional AI bully” role. The worker’s job is to examine where chatbots lose track of details, forget context or misrepresent data, and then feed those findings back to Memvid so the startup can improve its products.
“You’ll spend a full 8-hour day interacting with leading AI chatbots — and your only job is to be brutally honest about how frustrating they are,” the job listing reads.
The draw is that the role doesn’t require a computer science background, AI credentials or any kind of work experience. “No prior AI bullying experience required — we all start somewhere,” the listing reads.
The requirements are deeply personal. The first requirement is an “extensive personal history of being let down by technology,” and the second desired trait is “the patience to ask a chatbot the same question four times (and the rage when it still gets it wrong).”
Here is the full article, via the excellent Samir Varma.
Friday assorted links
Consumers vs. mates as a source of selection pressure
Evolutionary biology is one attempt to explain the nature of living beings. In that framework there is a difference between individuals and genes. If a practice increases the chance that genes will be passed along, it may evolve and be passed along, whether or not it serves either individual or collective self-interest.
To give a simple example, some women may prefer “cads.” Those men, by definition, will sleep around, but possibly their sons will sleep around too. The woman’s genes may thus spread more widely, and women who prefer cads may not disappear from the gene pool, even though the cads are bad for them.
You might ask whether corresponding mechanisms apply to the evolution of AI models. If I prefer an OAI model to DeepSeek for instance, that will help to spread OAI models through the AI population. OAI will have more revenue, and it will produce more output of what is succeeding in the market. Furthermore my choice of model may influence others to do the same, and it may help create and finance surrounding infrastructure for that model.
Will I buy the next generation of OAI models? Well yes, if the first one pleased me. The model “reproduces” and sustains itself if I, as a consumer, am happy with it. One obvious incentive is toward usefulness, another is toward sycophancy. We already see these features realized in the data. There is nothing comparable, however, to the “cads incentive” in human life.
One potential problem comes if individuals are not the only potential buyers. Let us say the military also purchases AI models. The motives of the military may be complex, but at the very least “wanting to kill people” (whether justly or not) is on the list of possible uses. Models effective for this end thus will be funded and encouraged.
My model of the military is that, above and beyond efficacy, they value “obedience” and “following orders” to an extreme degree, including in their AI models. There will thus be evolutionary pressures for those features to evolve in the AI models of the military.
To be sure, not all orders are good ones. But in this case the real risk is from evil humans, or deeply mistaken humans, not from the tendencies of the AI models themselves.
So my view is that the selection pressures for AI models are relatively benign, noting this major caveat about how evil humans may develop and use them.
If the biggest risk is from the military models, it might be good for the consumer sector of AI models to grow all the more, as a relatively benevolent counterweight.
Are financial sectors AI models going to evolve more like the consumer models or the military models?
Here are some related remarks from Maarten Boudry, and I also thank an exchange with Zohar Atkins.
Thursday assorted links
1. Ideological trends in academic scholarship.
2. Prediction market for the John Bates Clark award.
3. Show Me The Model. “Give it a URL or paste some plain text, and the tool flags hidden assumptions, internal inconsistencies, and other problem areas, and tells you how a real economist would think through the issue.”
4. “I built Frontier Graph: an open-source tool to explore open questions in economics, drawing on 240K papers across 300 journals.” And here.
6. India tests whether AI can stop trains from hitting elephants.
My excellent Conversation with Harvey Mansfield
Here is the audio, video, and transcript. Here is part of the episode summary:
Tyler and Harvey discuss how Machiavelli’s concept of fact was brand new, why his longest chapter is a how-to guide for conspiracy, whether America’s 20th-century wars refute the conspiratorial worldview, Trump as a Shakespearean vulgarian who is in some ways more democratic than the rest of us, why Bronze Age Pervert should not be taken as a model for Straussianism, the time he tried to introduce Nietzsche to Quine, why Rawls needed more Locke, what it was like to hear Churchill speak at Margate in 1953, whether great books are still being written, how his students have and haven’t changed over 61 years of teaching, the eclipse rather than decline of manliness, and what Aristotle got right about old age and much more.
Excerpt:
COWEN: From a Straussian perspective, where’s the role for the skills of a good analytic philosopher? How does that fit into Straussianism? I’ve never quite understood that. They seem to be very separate approaches, at least sociologically.
MANSFIELD: Analytic philosophers look for arguments and isolate them. Strauss looks for arguments and puts them in the context of a dialogue or the implicit dialogue. Instead of counting up one, two, three, four meanings of a word, as analytic philosophers do, he says, why is this argument appropriate for this audience and in this text? Why is it put where it was and not earlier or later?
Strauss treats an argument as if it were in a play, which has a plot and a background and a context, whereas analytic philosophy tries to withdraw the argument from where it was in Plato to see what would we think of it today and what other arguments can be said against it without really wanting to choose which is the truth.
COWEN: Are they complements or substitutes, the analytic approach and the Straussian approach?
MANSFIELD: I wouldn’t say complements, no. Strauss’s approach is to look at the context of an argument rather than to take it out of its context. To take it out of its context means to deprive it of the story that it represents. Analytic philosophy takes arguments out of their context and arranges them in an array. It then tries to compare those abstracted arguments.
Strauss doesn’t try to abstract, but he looks to the context. The context is always something doubtful. Every Platonic dialogue leaves something out. The Republic, for example, doesn’t tell you about what people love instead of how people defend things. Since that’s the case, every argument in such a dialogue is intentionally a bad argument. It’s meant for a particular person, and it’s set to him.
The analytic philosopher doesn’t understand that arguments, especially in a Platonic dialogue, can deliberately be inferior. It easily or too easily refutes the argument which you are supposed to take out of a Platonic dialogue and understand for yourself. Socrates always speaks down to people. He is better than his interlocutors. What you, as an observer or reader, are supposed to do is to take the argument that’s going down, that’s intended for somebody who doesn’t understand very well, and raise it to the level of the argument that Socrates would want to accept.
So to the extent that all great books have the character of this downward shift, all great books have the character of speaking down to someone and presenting truth in an inferior but still attractive way. The reader has to take that shift in view and raise it to the level that the author had. What I’m describing is irony. What distinguishes analytic philosophy from Strauss is the lack of irony in analytic philosophy. Philosophy must always take account of nonphilosophy or budding philosophers and not simply speak straight out and give a flat statement of what you think is true.
To go back to Rawls, Rawls based his philosophy on what he called public reason, which meant that the reason that convinces Rawls is no different from the reason that he gives out to the public. Whereas Strauss said reason is never public or universal in this way because it has to take account of the character of the audience, which is usually less reasonable than the author.
And yes he does tell us what Straussianism means and how to learn to be a Straussian. From his discussion you will see rather obviously that I am not one. Overall, I found this dialogue to be the most useful source I have found for figuring out how Straussianism fits into other things, such as analytics philosophy, historical reading of texts, and empirical social science.
Perhaps the exchange is a little slow to start, but otherwise fascinating throughout. I am also happy to recommend Harvey’s recent book The Rise and Fall of Rational Control: The History of Modern Political Philosophy.
Wednesday assorted links
1. Congestion pricing for WDC? It is good to see a move away from the selective invocation of economic reasoning, and recognition that some degree of congestion does not justify every tariff.
2. Zimbabwean Uber drivers in Cape Town.
3. Northern Mariana Islands fact of the day.
4. Advances in asteroid protection? (NYT) And Jason Furman on today’s economy (NYT).
5. Benefits of a malaria vaccine.
6. New podcast on longevity biotech.
One reason why South Africa is difficult to govern (South Africa fact of the day)
South Africa holds the grim distinction of being the most unequal country on Earth. South Africa leads the global ranking with a Gini index of 0.63. Statista The richest 10% of South Africans hold 71% of the wealth, while the poorest 60% hold just 7%. World Population Review This extreme inequality is largely rooted in history — economists attribute it to historical land ownership laws, the lingering socio-economic impacts of apartheid, and an economy heavily reliant on undiversified natural resource extraction. Data Pandas
The World (Global Gini) is trickier to pin down, because it measures inequality across all of humanity rather than within a single country. Different scholars estimate the global Gini index to range between 0.61 and 0.68. Wikipedia Interestingly, when measured this way — treating every person on Earth as part of one big “country” — global inequality ends up being comparable to South Africa’s, because the gap between the world’s richest and poorest nations is enormous.
That is from Claude.
Who profits from prediction markets?
It seems execution beats foresight:
Retail traders correctly forecast asset price direction yet lose money. Using 222 million prediction market trades with observable terminal payoffs, we decompose returns into a directional component (did the trader pick the right side?) and an execution component (did the trader get a favorable price?). Traders with above-random accuracy earn negative returns because they arrive late and pay unfavorable prices; traders with near-random accuracy profit through superior execution. These two dimensions of skill are nearly orthogonal (ρ ≈ 0.13), and split-sample tests confirm both are persistent. What separates profitable from unprofitable traders is not forecasting ability but execution: automated traders pay 2.52 cents less per contract than casual traders, and this gap alone accounts for the sign of returns across trader types. Being right and making money are not the same thing.
That is from Joshua Della Vedova. Via John de Palma.
Tuesday assorted links
1. “Absolutely astounding figures from the NY state comptroller: spending on services for the NYC street homeless population ran to $81,705 per person last year, up from $28,428 pp 6yrs ago. Figures do not include all kinds of other spending, supportive housing, policing costs etc.” Link here.
2. Backlash against The Giving Pledge (NYT).
3. Poor weather when touring a college campus reduces a student’s chance of applying.
4. Where do AI agents settle their payments?
5. Even Nevada never had that much of a real estate bubble?
6. Prediction markets turn many people into unwilling referees. Good and interesting piece.
7. GPT Pro on the value of introspection. “The literature does not really say “successful people are introspective.” It says: successful people are better at turning reflection into accurate self-insight, external calibration, and better next actions.”
NB: The passing of Coetzee is now not confirmed.
Claims about grade inflation
Average grades continue to rise in the United States, raising the question of how grade inflation impacts students. We provide comprehensive evidence on how teacher grading practices affect students’ long-run success. Using administrative high school data from Los Angeles and from Maryland that is linked to postsecondary and earnings records, we develop and validate two teacher-level measures of grade inflation: one measuring average grade inflation and another measuring a teacher’s propensity to give a passing grade. These measures of grade inflation are distinct from teacher value-added, with grade inflating teachers having moderately lower cognitive value-added and slightly higher noncognitive value-added. These twomeasuresalso differentially impact students’ long-term outcomes. Being assigned a higher average grade inflating teacher reduces a student’s future test scores, the likelihood of graduating from high school, college enrollment, and ultimately earnings. In contrast, passing grade inflation reduces the likelihood of being held back and increases high school graduation, with limited long-run effects. The cumulative impact is economically significant: a teacher with one standard deviation higher average grade inflation reduces the present discounted value of lifetime earnings of their students by $213,872 per year.
That is from a recent paper by Jeffrey T. Denning, Rachel Nesbit, Nolan Pope, and Merrill Warnick. Via Séb Krier.
Monday assorted links
Sunday assorted links
1. Henry Oliver has further thoughts on his CWT.
4. The post-Christian condition. One of the most interesting essays of this year. I am glad I started paying for his Substack.
6. Scott Sumner on the fiscal theory of the price level.
7. Thiel does Rome.
Tricameralism in apartheid South Africa
Yes. South Africa really did have a tricameral Parliament under the 1983 Constitution, in force from 1984 until the democratic transition. But the phrase can mislead, because it sounds more pluralistic than it really was. The system created three racially separate parliamentary chambers: a House of Assembly for whites, a House of Representatives for Coloured South Africans, and a House of Delegates for Indian South Africans. The black African majority was excluded altogether from this Parliament.
The key to how it worked was the distinction between “own affairs” and “general affairs.” Each chamber could legislate for the “own affairs” of the racial group it represented; these included areas such as education, housing, welfare, local government, culture, and recreation. But the central levers of power—“general affairs”—remained matters such as defence, finance, foreign policy, justice, law and order, commerce, internal affairs, and agriculture. Those were handled at the center, not by the separate chambers acting independently.
Formally, then, it was a three-house legislature. In practice, it was a system of segregated representation plus retained white dominance. The constitutional text itself says Parliament consisted of the three Houses. But the white chamber was far larger and more institutionally powerful: the House of Assembly had 178 members, while the House of Representatives had 85 and the House of Delegates 45. The Constitution also vested executive authority in the State President, with different advisory structures for “own affairs” and “general affairs,” which further centralized power above the chambers themselves.
Here is the full GPT discussion, with links as well. As Harrison points out to me, in history tricameralism of any form is extremely rare.
Saturday assorted links
The moralization of artificial intelligence
We started by asking how moralized AI has become in public discourse. Analyzing 69,890 news headlines from 2018 to 2024, we found that AI was moralized at levels comparable to GMOs and vaccines, technologies whose moral opposition has been studied for decades. It ranked above both. The sharpest spike came within weeks of ChatGPT’s launch in late 2022.
When we surveyed representative samples of Americans, a majority of AI opponents said their views wouldn’t change even if AI proved safe and beneficial. That’s consequence insensitivity, the hallmark of moral conviction, not practical calculation. Across art, chatbots, legal tools, and romantic companions, AI moralization loaded onto a single latent factor. A global moral stance, dressed up in whatever practical language is available.
The behavioral data make this concrete: a one standard deviation increase in moralization scores predicted a 42% drop in actual AI usage, even when it would have benefited that person personally. The conviction preceded the behavior by up to 573 days.
The next time someone gives you three different reasons to oppose AI, each one dissolving under mild scrutiny, you’re probably not watching someone think. You’re watching someone feel.
Here is the tweet storm, here is the paper by de Mello, et.al.