GPT-4
Tuesday assorted links
1. Open source LLMs? And more here. And running LLMs on consumer GPUs. And more.
2. Michael Jackson markets in everything.
3. Surveillance in the Danish welfare state (Wired).
4. Claim that the teen mental health dilemma started well before 2012.
5. LLMs vs. Chomsky. And more on Visual ChatGPT.
The roots of misinformation fears?
Alarmist narratives about the flow of misinformation and its negative consequences have gained traction in recent years. If these fears are to some extent warranted, the scientific literature suggests that many of them are exaggerated. Why are people so worried about misinformation? In two pre-registered surveys conducted in the United Kingdom (Nstudy_1 = 300, Nstudy_2 = 300) and replicated in the United States (Nstudy_1 = 302, Nstudy_2 = 299), we investigated the psychological factors associated with perceived danger of misinformation and how it contributes to the popularity of alarmist narratives on misinformation. We find that the strongest, and most reliable, predictor of perceived danger of misinformation is the third-person effect (i.e. the perception that others are more vulnerable to misinformation than the self) and, in particular, the belief that “distant” others (as opposed to family and friends) are vulnerable to misinformation. The belief that societal problems have simple solutions and clear causes was consistently, but weakly, associated with perceived danger of online misinformation. Other factors, like negative attitudes toward new technologies and higher sensitivity to threats, were inconsistently, and weakly, associated with perceived danger of online misinformation. Finally, we found that participants who report being more worried about misinformation are more willing to like and share alarmist narratives on misinformation. Our findings suggest that fears about misinformation tap into our tendency to view other people as gullible.
That is from new work by Sacha Altay and Alberto Acerbi, via the excellent Kevin Lewis.
Machine Learning as a Tool for Hypothesis Generation
While hypothesis testing is a highly formalized activity, hypothesis generation remains largely informal. We propose a systematic procedure to generate novel hypotheses about human behavior, which uses the capacity of machine learning algorithms to notice patterns people might not. We illustrate the procedure with a concrete application: judge decisions about who to jail. We begin with a striking fact: The defendant’s face alone matters greatly for the judge’s jailing decision. In fact, an algorithm given only the pixels in the defendant’s mugshot accounts for up to half of the predictable variation. We develop a procedure that allows human subjects to interact with this black-box algorithm to produce hypotheses about what in the face influences judge decisions. The procedure generates hypotheses that are both interpretable and novel: They are not explained by demographics (e.g. race) or existing psychology research; nor are they already known (even if tacitly) to people or even experts. Though these results are specific, our procedure is general. It provides a way to produce novel, interpretable hypotheses from any high-dimensional dataset (e.g. cell phones, satellites, online behavior, news headlines, corporate filings, and high-frequency time series). A central tenet of our paper is that hypothesis generation is in and of itself a valuable activity, and hope this encourages future work in this largely “pre-scientific” stage of science.
Here is the full NBER working paper by Jens Ludwig and Sendhil Mullainathan.
Sentences to ponder
Big banks’ behavior this time has been shaped by the fallout from 2008. Why isn’t Dimon buying S.V.B.? He has complained about the headaches of buying Bear Stearns and Washington Mutual at the government’s behest in 2008, having spent years fighting litigation and paying fines for those firms’ bad behavior. Bank executives who were around back then remember that.
That is from Dealbook 2.0, NYT, via TO. File under “The Costs of Intervention and Regulation and Political Grandstanding are Higher than You Think.”
Monday assorted links
1. Contrarian perspective on the current Israeli disputes.
2. Running cognitive pipelines on cheap hardware. And additive prompting.
3. Maxims. 17th century, Hansonian.
4. Turkey fact of the day: “Turkey borders seven different countries all of which use different alphabets.” Significant.
5. Machine learning as a tool for hypothesis generation.
6. “This sea slug cut off its own head — and lived to tell the tale.” And readjusting our expectations about the Fed, future rate hikes, and presumably inflation as well. Be ready people.
Can the SVB crisis be solved in the longer run?
The failure and closure of Silicon Valley Bank (SVB) raise immediate issues as to how policymakers should react. I’d like to step back and consider what this implies for banking regulation more generally in the longer run. The main lesson is that successful bank regulation is an ongoing, dynamic problem, unintended secondary consequences are rife, and neither more regulation or less regulation can be guaranteed to succeed.
If you think of the FDIC/Fed/Treasury as a consolidated entity, the broader question is how many financial institution liabilities they should guarantee, whether explicitly or implicitly. Let us consider why in fact the government felt compelled to guarantee all of the deposits.
An unwillingness to guarantee all the deposits would satisfy the desire to penalize businesses and banks for their mistakes, limit moral hazard, and limit the fiscal liabilities of the public sector. Those are common goals in these debates. Nonetheless unintended secondary consequences kick in, and the final results of that policy may not be as intended.
Once depositors are allowed to take losses, both individuals and institutions will adjust their deposit behavior, and they probably would do so relatively quickly. Smaller banks would receive many fewer deposits, and the giant “too big to fail” banks, such as JP Morgan, would receive many more deposits. Many people know that if depositors at an institution such as JP Morgan were allowed to take losses above 250k, the economy would come crashing down. The federal government would in some manner intervene – whether we like it or not – and depositors at the biggest banks would be protected.
In essence, we would end up centralizing much of our American and foreign capital in our “too big to fail” banks. That would make them all the more too big to fail. It also might boost financial sector concentration in undesirable ways.
To see the perversity of the actual result, we started off wanting to punish banks and depositors for their mistakes. We end up in a world where it is much harder to punish banks and depositors for their mistakes.
Another unintended secondary consequence is that lots of funds would flow out of the banking system and into U.S. Treasuries. In other words, our private businesses would find it harder to borrow and our government would find it easier to borrow and thus government would command more resources. That hardly seems like a desirable outcome for a policy decision that had some initially libertarian motives.
Alternatively, you might think it is a simple way out for the government to guarantee all those deposits, as indeed was done last evening.
That decision too will prove to have unintended secondary consequences. Raising the FDIC protection limit from $250,000 to ??? raises political eyebrows in a dramatic manner. For one thing, the FDIC would then be seen as guaranteeing a much larger part of the financial system. Over time, the pressures for the government to protect yet additional parts of the financial system will grow, just as they did after the bailouts from the 2008-2009 financial crisis. Furthermore, if the FDIC keeps on increasing its protections in the quest for financial stability, that means a larger FDIC, a larger regulatory apparatus, perhaps higher capital requirements, and over time higher premia for banks to pay to the FDIC. (As a side note, worthy of another post, we are also hearing calls that somehow VCs need to be regulated now, if they are going to “receive bailouts”.)
As that scenario unfolds, there will be all the more incentives to supply more lending and also deposit-taking services outside the formal and more heavily regulated banking sector. Rather than pushing more resources into the larger banks, this policy would push additional resources outside the formal banking system altogether. That would mean less power, oversight and scrutiny from the Fed and also from other regulatory bodies. Typically American banks are more tightly regulated and monitored than are non-bank financial entities.
This kind of problem is likely to unfold slowly, but it is no less real. The initial policy was an expansion of FDIC regulatory authority, but the end result could well be less total regulation of lending and depository functions. Once again, the policy decision may fail at achieving its initially intended goals.
The core problem is this: regulators can only protect so much of the financial system. Yet in a wealthy, peaceful economy the financial system often grows more rapidly than does gdp, if only because the financial system is based on the intermediation of wealth, not income. Simple accumulation boosts the ratio of wealth to income over time, thereby creating regulatory dilemmas for finance. Neither “regulating more and more of it” nor “letting more and more of it continue in a less regulated manner” are entirely satisfactory solutions.
But those of course are the only options available to us.
“Authorities Reinstate Alcohol Ban for Aboriginal Australians”
Geoff Shaw cracked open a beer, savoring the simple freedom of having a drink on his porch on a sweltering Saturday morning in mid-February in Australia’s remote Northern Territory.
“For 15 years, I couldn’t buy a beer,” said Mr. Shaw, a 77-year-old Aboriginal elder in Alice Springs, the territory’s third-largest town. “I’m a Vietnam veteran, and I couldn’t even buy a beer.”
Mr. Shaw lives in what the government has deemed a “prescribed area,” an Aboriginal town camp where from 2007 until last year it was illegal to possess alcohol, part of a set of extraordinary race-based interventions into the lives of Indigenous Australians.
Last July, the Northern Territory let the alcohol ban expire for hundreds of Aboriginal communities, calling it racist. But little had been done in the intervening years to address the communities’ severe underlying disadvantage. Once alcohol flowed again, there was an explosion of crime in Alice Springs widely attributed to Aboriginal people.
Here is more from Yan Zhuang at the New York Times. Via Rich Dewey.
That was then, this is now
From Taylor C. Sherman’s useful Nehru’s India: A History in Seven Myths:
Although Hindu nationalists had gained prominence in the run-up to partition, the new Congress leaders of the Government of India tried to sideline them. After Gandhi’s assassination on 30 January 1948, members of the Rashtriya Swayamsevak Sangh were arrested, and the Hindu Mahasabha declared it would not take part in politics. In short, though raging before partition, the flames of Hindu chauvinism were quickly doused after independence, at least according to the old nationalist narrative. Secondly, the reform of Hinduism was seen as an essential element of secularism. To this end, a prominent Dalit, Bhimrao Ramji Ambedkar, was put in charge of both writing the Constitution and overseeing reform of Hindu personal law. Within a short time after independence, so the myth goes, India had a secular state, and was on course to establish a sense of security and belonging for the two groups who had raised the loudest objections to Congress’s nationalism: Muslims and Dalits.
As with so many of the myths that have arisen about this period after independence, the myth of India secularism owes a great deal to Jawaharlal Nehru.
The book is both a good focused view of the Nehru era, but excellent background for current disputes.
Sunday assorted links
1. The banking culture that is Greek-Nigerian. And why so many start-up people used SVB.
2. Visual ChatGPT. And Ezra is coming around on AI (NYT).
3. The Instagram face, and American class.
4. The unusual article that sits in the narrow Venn diagram between “would pass for ridiculous satire” and “nonetheless is almost entirely true.”
What I’ve been reading
1. Tezer Özlü, Cold Nights of Childhood. A Turkish novella, originally published in 1980, newly translated into English and the first English-language book by her. I give this one an A/A+, mostly emotional drama and narrative, I can’t tell you more without spoilers. Here is more on the author. Only 76 pp. When will her suicide book be published in English?
2. Jens Heycke, Out of the Melting Pot, Into the Fire: Multiculturalism and the World’s Past and America’s Future. Argues that ethnic divisions should be made less rather than more focal: “When I visited Rwanda, I asked Rwandans of various backgrounds whether they thought distinguishing people by race or ethnicity ever helped anyone in their country.” An effective presentation of facts, though only one side of the story and it does not take sufficiently seriously the question of how tolerant environments ever get established in the first place (hint: it is through a certain amount of identity politics…what exactly is an Englishman anyway?).
3. Slavoj Žižek and John Milbank, The Monstrosity of Christ: Paradox or Dialectic? How Hegelian should our understanding of Christ be? The book is written as a confrontational dialogue, and to its benefit. You do need to be able to stomach sentences such as: “Do the three main versions of Christianity not form a kind of Hegelian triad?” (SZ) In any case, the smartness of the authors makes it worthwhile. Once you move past their immediate (and extreme) fan bases, both are in fact considerably underrated.
There is Peter Baldwin’s Athena Unbound: Why and How Scholarly Knowledge Should be Free for All.
And Nicolas Spencer, Magisteria: The Entangled Histories of Science and Religion.
Ashoka Mody has published the quite pessimistic India is Broken: A People Betrayed, Independence to Today.
Plus quite a few others that I don’t feel the need to tell you about…
Inflation is slightly underrated
Still bad, yes, but it has a few upsides, here is one part from my Bloomberg column:
It’s also worth asking which groups are most hurt by inflation, and which least. On the one hand, inflation helps debtors, and the debtor class is often relatively poor. Yet there is also a large set of distributional effects through wages, and those effects might prove more congenial to conservative values.
One obvious loser from higher inflation is someone with a tenured job, either de facto or de jure. To use an example close to home: Many professors cannot easily switch jobs and duplicate their current working conditions, nor can they easily demand and receive offsetting raises. A lot of them are locked into their current posts — which is hardly surprising, since the nature and incentives of tenure do not require those who have it to stay at the top of their game to keep their job.
Many conservatives wish to abolish academic tenure altogether. That’s unlikely to happen (though Florida is making moves in that direction), but a dose of inflation might make tenure less appealing.
Many government bureaucrats also often have de facto tenure, and many of them are not so marketable to the private sector. For them, higher inflation is a pay cut. Again, just as inflation will not abolish tenure, so it will not eliminate bureaucracy. But it may make the less dynamic parts of the public sector less attractive, helping to slow the growing bureaucratization of society.
In contrast, consider a private entrepreneur who will attempt seven start-ups over the next 20 years, with some succeeding and others failing. That person probably will not suffer much from inflation, as he is starting from scratch with new nominal values on a fairly frequent basis. More generally, productive people who have flexibility and the ability to adjust to circumstances will emerge relatively unscathed as well.
Here is some commentary from Scott Sumner, though I think Scott is not taking me literally enough. Inflation really is bad (“never reason from an inflation change!”…nonetheless), it is simply elsewhere that I explain my views on that. And do note an extensive literature shows that the Fisher effect does not hold 1-to-1, and thus inflation at modest levels does lower real borrowing rates. It is also funny for Scott (“Money Illusion”) to say that short-run non-neutralities do not matter.
Are smarter entities more coherent?
There is an assumption behind this misalignment fear, which is that a superintelligent AI will also be supercoherent in its behavior1. An AI could be misaligned because it narrowly pursues the wrong goal (supercoherence). An AI could also be misaligned because it acts in ways that don’t pursue any consistent goal (incoherence). Humans — apparently the smartest creatures on the planet — are often incoherent. We are a hot mess of inconsistent, self-undermining, irrational behavior, with objectives that change over time. Most work on AGI misalignment risk assumes that, unlike us, smart AI will not be a hot mess.
In this post, I experimentally probe the relationship between intelligence and coherence in animals, people, human organizations, and machine learning models. The results suggest that as entities become smarter, they tend to become less, rather than more, coherent. This suggests that superhuman pursuit of a misaligned goal is not a likely outcome of creating AGI.
That is from a new essay by Jascha Sohl-Dickstein, speculative but interesting. Via N.
Saturday assorted links
1. The (new) Krugman version of the sectoral shifts hypothesis (NYT).
2. My short essay “Do Non-profits Drive Social Change?”
3. Africa fact of the day: “Today there are more city dwellers in Kenya than there were in all of Africa in 1950”
4. Robots performing Hindu rituals.
5. AI-generated Django (sort of), note the rapid improvement.
6. Use cases for ChatGPT (long). And how to use Chat GPT to give better public talks.
Which political tweets do best?
Analyzing every tweet of all US senators holding office from 2013 to 2021 (861,104 tweets from 140 senators), we identify a psycholinguistic factor, greed communication [TC: basically accusing other people of greed], that robustly predicts increased approval (favorites) and reach (retweets). These effects persist when tested against diverse established psycholinguistic predictors of political content dissemination on social media and various other psycholinguistic variables. We further find that greed communication in the tweets of Democratic senators is associated with greater approval and retweeting compared to greed communication in the tweets of Republican senators, especially when those tweets also mention political outgroups.
That is from new research by Eric J. Mercadante, Jessica L. Tracy, and Friedrich M. Götz. Via David Lilienfeld.