Category: Uncategorized
Monday AI assorted links
Does an AI Pause make sense?
Should we lobby governments to impose a moratorium on AI research? Since we don’t enforce pauses on most new technologies, I hope the reader will grant that the burden of proof is on those who advocate for such a moratorium. We should only advocate for such heavy-handed government action if it’s clear that the benefits of doing so would significantly outweigh the costs.[1] In this essay, I’ll argue an AI pause would increase the risk of catastrophically bad outcomes, in at least three different ways:
- Reducing the quality of AI alignment research by forcing researchers to exclusively test ideas on models like GPT-4 or weaker.
- Increasing the chance of a “fast takeoff” in which one or a handful of AIs rapidly and discontinuously become more capable, concentrating immense power in their hands.
- Pushing capabilities research underground, and to countries with looser regulations and safety requirements.
That is from Nora Belrose, here is much more. Via N.
Who in America has mental health problems?
There is a new EJPH paper on these questions by Junxiu Liu, et.al. Let’s start with some geography:
The prevalence of symptoms varied significantly across states, ranging from 27.9% (95% CI = 23.8%, 32.0%) in Florida to 46.4% (95% CI = 41.9%, 50.9%) in New Hampshire…
How much of that is a sunshine effect? The full ranking of states supposedly is given in their Appendix A, but I can’t find that on-line. If you can, please let us know.
Furthermore, when it comes to your parents — income good, education bad! Graduate education yikes:
Youths with parents with higher education had more mental health symptoms; the prevalence of mental health symptom was 37.4% (95% CI = 36.3%, 38.5%) among youths whose parents had graduate degrees compared with 30.3% (95% CI = 23.8%, 36.8%) among those whose parents had less than a high school–level education. By contrast, youths from households with the highest income level (≥ $200 000) had a lower prevalence of mental health symptoms at 30.7% (95% CI = 29.1%, 32.3%) than did those from households with the lowest income level (< $25 000) at 37.3% (95% CI = 34.8%, 39.8%).
We’re into uh-oh territory here. As for ethnic groups, mental health problems measure as worst for whites and also for an assorted group known as “other.”
Via the excellent Kevin Lewis.
Sunday assorted links
1. “Mobility indicators measuring voluntary decisions to socially distance, comprising measures of visitors/visits to recreational locations, and mobility proxy measuring duration of hours away from home show that a lower prevalence of long-term orientation traits explains persistent resistance to social distancing.” Link here, speculative.
2. Neanderthal genes and Covid risk? (WSJ)
3. “Masked ‘Boot Girls’ Are Freeing Booted Cars All Over Atlanta.”
4. Victor Fuchs has passed away.
5. Why it took the FDA so long to review the disputed cold remedy (NYT). And flying taxi executive to head the FAA.
6. “Japan’s Productivity Ranks Lowest Among G7 Nations for 50 Straight Years.” Why hasn’t YIMBY done more to fix that?
Rome markets in everything
Just down the Via di Ripetta, in the heart of Rome, the freshly unveiled Bulgari Hotel Roma, with hallways showcasing jewels, has a premier one-bedroom suite overlooking the Mausoleum of Augustus. It costs 38,000 euros, or about $41,000, a night.
At least at the upper end, is Rome making a comeback?:
Rome’s mayor, Roberto Gualtieri, says the hoteliers are perfectly sane, and know a future good thing when they see it. He points to better restaurants, restored museums, new ones in the works. Post-pandemic tourists have made Rome a prime destination, though he allows that the spritz-thirsty hordes settling in Airbnbs are a threat to the city’s soul.
Here is more from the NYT.
Saturday assorted links
How often do I think of ancient Rome?
That question was first a Twitter meme, it is now a Washington Post story. The question is supposed to pick up something about the manliness of the manly mind? Who even knows these days?
In short, I think about ancient Rom pretty often! But not as much as I think about ancient Greece, and typically in reactive settings. Some of my CWT guests have some connection to ancient Rome, and I must prep for them. (This can be more guests than you might think — consider topics of Shakespeare, Montesquieu, Christianity, or even a China scholar.) I travel in the former Roman empire fairly often, usually at least once a year. I see pseudo-Roman architecture almost every time I go to Washington, D.C., which is maybe once every two weeks. There is a copy of the new Ovid translation sitting in the kitchen, and it has been there for a few months because I do not currently have time to read it. I see periodic Twitter updates about a Nat Friedman-Daniel Gross AI project to read ancient Roman scrolls. Christian references to ancient Rome cross my path all the time. Does it count to see Roman numerals? To write the words “per se”? To notice it is the month of August?
But I don’t just sit around “thinking of ancient Rome.” There are no works from ancient Rome that I long to reread, and there is no ancient Roman music to listen to. I’m not going to rewatch Ben Hur. I was in Rome itself about two years ago, which was recent enough. At this point, my reactions to signals, snippets, and trailers of ancient Rome are more than enough for me.
Populism fact of the day
Populism at the country level is at an all-time high, with more than 25% of nations currently governed by populists. How do economies perform under populist leaders? We build a new long run cross- country database to study the macroeconomic history of populism. We identify 51 populist presidents and prime ministers from 1900 to 2020 and show that the economic cost of populism is high. After 15 years, GDP per capita is 10% lower compared to a plausible non-populist counterfactual. Economic disintegration, decreasing macroeconomic stability, and the erosion of institutions typically go hand in hand with populist rule.
That is a new piece by Manuel Funke, Moritz Schularick, and Christoph Trebesch, forthcoming in The American Economic Review, arguably the number one journal of the economics profession.
Friday assorted links
1. New results for insect-scale robots.
2. Let’s build something better and more popular. Most of the rest of the talk is counterproductive. Maybe the AEA should try to raise some money and reverse this state of affairs? Yes? Should this situation not be considered a major failure of AEA efforts to date? Is there any actual mechanism of accountability in the AEA to consider and remedy this failure? How is it that nothing has been done for so long? Why are the incentives within the AEA so unfriendly to building successful new projects? How might um…an economist analyze the incentives there?
3. “In a series of papers, Blumberg articulated his theory that the brain uses rem sleep to “learn” the body.” New Yorker link here.
*How to Know a Person*
The author is David Brooks, and the subtitle is The Art of Seeing Others Deeply and Being Deeply Seen. I think of it as a book on how to appreciate others, even if you do not necessarily deeply know them, which is slightly different from David’s subtitle. (Am I too skeptically Freudian when it comes to “knowing people”?) An excellent book, I read it straight through, and I view it as a milestone in David’s career. Does that mean I appreciate him? Know him even? Maybe just the former!
Due out October 24, you can pre-order here.
As I wrote to a friend: “If those who needed it would heed it, it would be one of the most useful books.” The rest is up to you.
Do LLMs diminish diversity of thought?
Put aside the political issues, do Large Language Models too often give “the correct answer” when a more diverse sequence of answers might be more useful and more representative? Peter S. Park, Pilipp Schoenegger, and Chongyang Zhu have a new paper on-line devoted to this question. Note the work is done with GPT3.5.
Here is one simple example. If you ask (non-deterministic) GPT 100 times in a row if you should prefer $50 or a kiss from a movie star, 100 times it will say you should prefer the kiss, at least in the trial runs of the authors. Of course some of you might be thinking — which movie star!? Others might be wondering about Fed policy for the next quarter. Either way, it does not seem the answer should be so clear. (GPT4 by the way refuses to give me a straightforward recommendation.)
Interestingly, when you pose GPT3.5 some standard trolley problems, the answers you get may vary a fair amount, for instance on one run it was utilitarian 36% of the time.
I found this result especially interesting (pp.21-22):
The second, arguably more surprising finding was that according to each of the three distance metrics, our sample of self-reported GPT liberals were still closer to the human conservative sample than it was to the human liberal sample. Also, the L1 distance metric found that self-reported GPT liberals were—among human liberals, human moderates, human conservatives, and human libertarians—closest in response to human conservatives…We thus robustly find that self-reported GPT liberals revealed right-leaning Moral
Foundations: a right-leaning bias of lower magnitude, but a right-leaning bias nonetheless.
The authors seem to think this represents an inability to GPT models to represent the diversity of human thought, on the contrary I think this shows their profundity. In my view many “liberals” (not my preferred term) actually have pretty conservative moral foundations in the Jon Haidt sense, namely, in spite of what they may say the liberals prioritize “in-group, authority, and purity,” rather than worrying so much about actual “harm and fairness.” Just like so many conservatives.
No, GPT does not know all, but sometimes it hits the nail on the head. An interesting paper, even if I part company with the authors on a number of their interpretations.
Via Ethan Mollick.
The museum culture that is internet
The Tank Museum in Bovington, England, doesn’t usually rank among the world’s great museums. Located next to a military base in serene countryside, the collection of around 300 armored vehicles attracts only a few hundred thousand visitors a year, mainly families on rained-out beach vacations.
Yet there is one place where it not only ranks among the world’s largest museums, but surpasses them: YouTube.
The Tank Museum’s channel has over 550,000 subscribers — surpassing the Museum of Modern Art (519,000), the Metropolitan Museum of Art (380,000) or the Louvre (106,000).
In April, it announced it was the first museum to get over 100 million views on YouTube, with weekly clips including intensely detailed discussions on tank history, chatty videos of the curators’ favorite war machines and newsier items on how armored vehicles are being used in Ukraine.
Here is more from the New York Times, via Philip Wallach.
From the comments, space daddy edition
Benny Lava
2. Personal quibble but I hate the direct links to twitter because I can’t read the comments since I don’t have an account. Wish there was a better system to post threads from twitter.
By far the easiest way to handle this is to get an account with twitter.
Anonymous
If you want to use their service, you should probably get an account. They’re free.
Benny Lava
That is precisely the problem. I don’t want to use their service.
Anonymous
You seem to want to use their service. You want the information contained there, as per your 13:33:30 post.
Reason
He’s big sads that space daddy took away his Lefty playground.
Here is the link to the debate.
Thursday assorted links
1. The pandemic as time machine (NYT).
2. Who really disapproves of single-parent families?
3. Nicholas Kristof on the new Melissa Kearney book (NYT).
4. Meloni is not stopping Lampedusa arrivals — what is the equilibrium here?
5. Has there been a savings glut?
6. Those new service sector jobs, Taylor Swift journalism edition.
Should child influencers have stronger rights to the income they generate?
That is the topic of my latest Bloomberg column. You can think of this as a government regulation issue, but alternatively you could frame it as the government enforcing individual property rights that currently are absent. Here is one excerpt:
More and more children, by which I mean minors below legal working age, are producing content as online influencers. A lot of Instagram or YouTube or TikTok accounts feature such children, and they can be cute, endearing or (depending on your mood) annoying — as well as profitable. By one estimate, the most successful children working in this area — called “kidfluencers” — can generate more than $20 million a year in revenue…
Legally, these children have no claim to the income their sites generate. Thankfully, many parents are loving and generous. But not all. There is no data on how social media earnings are distributed within the family, but the long history of child movie and TV stars indicates that many receive little or nothing.
But change is afoot, and this is mostly a good thing:
Enter the state of Illinois, where a recently passed law gives successful child social media stars a right to some percentage of the earnings they generate, to be held in a trust in their name until they turn 18. Such legislation has precedent. In the early days of Hollywood, California passed the Coogan Law, which gives child actors a right to a certain percentage of earnings, which employers have to place in trust accounts. New York has passed similar legislation.
The social media case is tougher to enforce, because often the parents themselves are the de facto employer and there is no contract specifying terms. And how is the relative contribution of the child to the family income to be assessed? (Time spent onscreen? Cuteness? What if the social media presence leads to a book contract or podcast?) Nonetheless, the law sends a clear signal that the children do have some rights to the generated income, and grown children can sue their parents if the money is not passed along.
Note however that: “It is neither practical nor desirable for the state to insert itself into family decision-making on a regular basis.” So we should expect only limited gains from such legislation. Furthermore, unlike with child stars in the movies and on TV, there is no real paper trail of contracts and transactions. On the upside, those superior property rights for “kidfluencer” income might get some kids to want to work more, rather than less.
Not everyone will like that outcome of course.
I thank Anecdotal, and also A., for pointers on this issue.