Results for “age of em” 15635 found
Approaching Human-Level Forecasting with Language Models
Forecasting future events is important for policy and decision making. In this work, we study whether language models (LMs) can forecast at the level of competitive human forecasters. Towards this goal, we develop a retrieval-augmented LM system designed to automatically search for relevant information, generate forecasts, and aggregate predictions. To facilitate our study, we collect a large dataset of questions from competitive forecasting platforms. Under a test set published after the knowledge cut-offs of our LMs, we evaluate the end-to-end performance of our system against the aggregates of human forecasts. On average, the system nears the crowd aggregate of competitive forecasters, and in some settings surpasses it. Our work suggests that using LMs to forecast the future could provide accurate predictions at scale and help to inform institutional decision making.
That is from a new paper by Danny Halawi, Fred Zhang, Chen Yueh-Han, and Jacob Steinhardt. I hope you are all investing in that chrisma…
Average is Over, love and romance edition
The dating site Tinder reports that, in 2023, 41 percent of Gen Z users were open to or seeking non-monogamous relationships, and 26 percent were open to ‘hierarchical polyamory’.
Here is the link. On the AI side, from the London Times:
What all my AI girlfriends have in common is that they foster pseudo-intimacy at lightning speed, they all have extra cost levels and they all “gamify” the experience somehow, whether through collectable badges, gems or levels of achievement which reward interaction with the AI.
Good luck young ones, what is the (partially) offsetting change in norms this will produce?
On-line images and lookism
Each year, people spend less time reading and more time viewing images1, which are proliferating online. Images from platforms such as Google and Wikipedia are downloaded by millions every day, and millions more are interacting through social media, such as Instagram and TikTok, that primarily consist of exchanging visual content. In parallel, news agencies and digital advertisers are increasingly capturing attention online through the use of images, which people process more quickly, implicitly and memorably than text. Here we show that the rise of images online significantly exacerbates gender bias, both in its statistical prevalence and its psychological impact. We examine the gender associations of 3,495 social categories (such as ‘nurse’ or ‘banker’) in more than one million images from Google, Wikipedia and Internet Movie Database (IMDb), and in billions of words from these platforms. We find that gender bias is consistently more prevalent in images than text for both female- and male-typed categories. We also show that the documented underrepresentation of women online is substantially worse in images than in text, public opinion and US census data. Finally, we conducted a nationally representative, preregistered experiment that shows that googling for images rather than textual descriptions of occupations amplifies gender bias in participants’ beliefs. Addressing the societal effect of this large-scale shift towards visual communication will be essential for developing a fair and inclusive future for the internet.
That is from a new Nature paper by Douglas Guilbeault, Solène Delecourt, Tasker Hull, Bhargav Srinivasa Desikan, Mark Chu, and Ethan Nadler. In general, print is much more gender-egalitarian than is images. Via the excellent Kevin Lewis.
A periodic reminder of your pending competitive inadequacy
Many people think “I will do […], AI will not anytime soon do [….] as well as I will.” That may or may not be true.
But keep in mind many of us are locked into a competition for attention. AI can beat you without competing against you in your task directly. What AI produces simply might draw away lots of attention from what you hope to be producing. Maybe looking Midjourney images, or chatting with GPT, will be more fun than reading your next column or book. Maybe talking with your deceased cousin will grip you more than the marginal new podcast, and so on.
This competition can occur even in the physical world. There will be many new, AI-generated and AI-supported projects, and they will bid for real resources. How about “AI figures out cost-effective desalination and so many deserts are settled and built out”? That will draw away resources from competing deployments, and your project will have to bid against that.
I hope it’s good.
Nash’s Contributions to Mathematics
Nash won the Nobel prize in Economics for his 2-page proof of Nash equilibrium, among the slightest of his achievements. Nash’s truly staggering contributions were in his embedding theorems, according to Gromov “one of the main achievements of mathematics of the twentieth century”. In this excellent talk, Cédric Villani gives an accessible guide to these theorems for mere mortals. Villani is a Fields medal winner, a French politician, and a character, all of which adds to the talk.
What should I ask Coleman Hughes?
I will be doing a Conversation with him, based in part around his new book The End of Race Politics: Arguments for a Colorblind America. On Coleman more generally, here is Wikipedia:
Coleman Cruz Hughes (born February 25, 1996) is an American writer and podcast host. He was a fellow at the Manhattan Institute for Policy Research and a fellow and contributing editor at their City Journal, and he is the host of the podcast Conversations with Coleman.
Also from Wikipedia:
Hughes began studying violin at age three. He is a hobbyist rapper—in 2021 and 2022, he released several rap singles on YouTube and Spotify, using the moniker COLDXMAN, including a music video for a track titled “Blasphemy”, which appeared in January 2022. Hughes also plays jazz trombone with a Charles Mingus tribute band that plays regularly at the Jazz Standard in New York City.
I saw Coleman perform quite recently, and I can vouch for his musical excellence, including as a singer. So what should I ask Coleman?
One reason why the disinflation proved so manageable
From a new and important paper by Xiwen Bai, Jesús Fernández-Villaverde, Yiliang Li, and Francesco Zanetti:
Our analysis shows that supply chain disruptions generate stagflation, accompanied by an increase in spare capacity for producers. This higher spare capacity curtails the supply of goods to retailers and results in a surge in prices, leading to a tighter retail market. We show that, in this situation, prices become highly sensitive to changes in demand, while output remains relatively inelastic. In other words, disruptions to the supply chain enhance the effectiveness of contractionary monetary policy in taming inflation while reducing the sensitivity of output to the policy. Our results reinforce the general findings on the state-dependence of the efficacy of monetary policy.
Progress! And arguably this is a Keynesian result:
In fact, our result resembles the celebrated analysis by Keynes (1940). Keynes argued that when output is constrained (in our case, because of supply chain disruptions, in Britain’s case in 1940, because of resources employed in World War II), policymakers can lower aggregate demand aggressively to prevent inflation without much fear of lowering production.
As I mention in GOAT, How to Pay for the War remains a neglected and underrated work by Keynes. Note this as well:
We document how supply chain shocks drove inflation during 2021 but that, in 2022, traditional demand and supply shocks also played an important role in explaining inflation.
The pure supply-side story, as you have been hearing from Krugman just isn’t going to work, it is time to give it up.
By the way, yet another advance in this lengthy paper is this (from JFV): “…we have satellite information about every single container ship of the world in real time (ID, geolocation, speed, draft, heading,…), which allows us to do tons of things in terms of machine learning and time series econometrics that people have not been able to do before.”
Comparing Large Language Models Against Lawyers
This paper presents a groundbreaking comparison between Large Language Models and traditional legal contract reviewers, Junior Lawyers and Legal Process Outsourcers. We dissect whether LLMs can outperform humans in accuracy, speed, and cost efficiency during contract review. Our empirical analysis benchmarks LLMs against a ground truth set by Senior Lawyers, uncovering that advanced models match or exceed human accuracy in determining legal issues. In speed, LLMs complete reviews in mere seconds, eclipsing the hours required by their human counterparts. Cost wise, LLMs operate at a fraction of the price, offering a staggering 99.97 percent reduction in cost over traditional methods. These results are not just statistics, they signal a seismic shift in legal practice. LLMs stand poised to disrupt the legal industry, enhancing accessibility and efficiency of legal services. Our research asserts that the era of LLM dominance in legal contract review is upon us, challenging the status quo and calling for a reimagined future of legal workflows.
That is from a new paper by Lauren Martin, Nick Whitehouse, Stephanie Yiu, Lizzie Catterson, and Rivindu Perera. Via Malinga.
Nominal wages are sticky in prison
“The only problem I have with my job is the pay hasn’t increased in over two decades,” wrote Nelson. At the time, she worked in a prison factory where she was forced to save two weeks’ pay “to buy one 18-pack of feminine hygiene pads,” a basic necessity.
Here is the full story, via The Browser.
Student Demand and the Supply of College Courses
From a recent Jacob Light paper:
In an era of rapid technological and social change, do universities adapt enough to play their important role in creating knowledge? To examine university adaptation, I extracted the information contained in the course catalogs of over 450 US universities spanning two decades (2000-2022). When there are changes in student demand, universities respond inelastically, both in terms of course quantity and content. Supply inelasticity is especially pronounced in fields experiencing declining demand and is more pronounced at public universities. Using Natural Language Processing, I further show that while the content of existing courses remains largely unchanged, newly-created courses incorporate topics related to current events and job skills. Notably, at selective institutions, new content focuses on societal issues, while at less selective institutions, new content emphasizes job-relevant skills. This study contributes uniquely to our understanding of the supply-side factors that affect how universities adapt to the rapidly evolving landscape.
John Cochrane offers comment as well, the first half of the post is interesting on demographics also.
Social improvements that don’t create countervailing negative forces
Let us say you favor policy X, and take steps to see that policy X comes about.
Under many conditions, people who favor non-X will take additional countervailing steps to oppose X. And in that case your actions in favor of X, on average, will lead to nothing. In the meantime, you and also your opponents will have wasted material resources fighting over X.
This argument is hardly new, but most people do not like to consider it much. They instead prefer to mood affiliate in favor of X, or perhaps against X. They prefer to be “fighting for the right things.”
Perhaps visible political organizing is most likely to set this dynamic in motion. Everyone can see what you are doing, and perhaps they can use their actions to fundraise for their own side.
That is one reason why I am not so thrilled with much of that organizing, even if I agree with it. Of course there are other scenarios here. Your involvement on behalf of X might just be flat-out decisive. Or perhaps the group against X is too resource-constrained to respond to your greater advocacy. That said, those descriptors (and others) might apply as well to either side of the dispute, your side included. Scaling up the fight over X might cause you to be the one who simply flat out loses the struggle.
It is worth thinking which kinds of “small steps toward a much better world” do not produce such countervailing effects.
How about “being positive and constructive”? Does it generate an equal and offsetting amount of negativity?
How about “trying to get people to be more reasonable, yet without offering a substantive political commitment bundled with that”? Does that in turn motivate the crazies to work harder at making everyone go insane? I am not sure.
What else might be effective, once these strategies are considered?
Does “refuting people” fit into this category? Yes or no?
Which activities should you be abandoning altogether? Or perhaps trying to do in secret, rather than publicly?
Settembrini and the continuing relevance of classical liberalism
Adrian Wooldridge has an excellent Bloomberg column on this topic, promoting the relevance of Thomas Mann, and here is one excerpt:
In the book [Magic Mountain], Castorp falls in with two intellectuals who live in the village of Davos below his sanitorium: an Italian humanist called Lodovico Settembrini and a Jewish-born cosmopolitan called Leo Naphta who is drawn to the Communist revolution and traditional Catholicism. The two men carry on a bitter argument about the relative merits of liberalism and illiberalism that touches on every question that mattered in prewar Europe: nationalism, individualism, fairness, tradition, war, peace, terrorism and so on.
Settembrini mechanically repeats the central tenets of liberalism but doesn’t seem to realize that the world is a very different place from what it was in 1850…Settembrini is like the bulk of today’s liberals — well-meaning but incapable of recognizing that the world of their youth has changed beyond recognition.
My reading of the world, however, is slightly different. I think the Settembrini example, from 1924, shows classical liberalism is still relevant. In 1924, classical liberalism seemed out of touch because the rest of the world was too fascistic, too communist, and too negative, among other problems. Yet at the time the classical liberals were essentially correct, even though Settembrini sounds out of touch.
Because classical liberals continued to carry the torch, we later had another highly successful classical liberal period, something like 1980-2000, though of course you can argue the exact dates.
Perhaps the underlying model is this: classical liberals often seem out of touch, because the world is too negative to respond to their concerns. Most of the time classical liberals are shouting into the well, so to speak. But they need to keep at it. Every now and then a window for liberal change opens, and then the classical liberals have to be ready, which in turn entails many years in the intellectual and ideological wilderness.
When the chaos surrounds, the liberals are no less relevant. The Settembrini character, from 1924, illustrates exactly that. Because he did eventually have his day, though many years later.
A congestion theory of unemployment fluctuations
Yusuf Mercan, Benjamin Schoefer, and Petr Sedláček, newly published in American Economic Journal: Macroeconomics. I best liked this excerpt from p.2, noting that “DMP” refers to the Nobel-winning Diamond-Mortensen-Pissarides search model of unemployment:
This congestion mechanism improves the business cycle performance of the DMP model considerably. It raises the volatility of labor market tightness tenfold, to empirically realistic levels. It produces a realistic Beveridge curve despite countercyclical separations. On its own, it accounts for around 30–40 percent of US unemployment fluctuations and much of its persistence. In addition, the model accounts for a range of other business cycle patterns linked to unemployment: the excess procyclicality of wages of newly hired workers compared to average wages, the countercyclical labor wedge, large countercyclical earnings losses from displacement and from labor market entry, and the long-run insensitivity of unemployment to policies such as unemployment insurance.
And by their congestion mechanism the authors mean this:
…a constant returns to scale aggregate production function that exhibits diminishing returns to new hires, a feature we call congestion in hiring.
I find that assumption plausible. It remains the case that the DMP model is grossly underrepresented in on-line writings on economics, on Twitter, and in the blogosphere. It won three Nobel Prizes, yet it also suggests that the “simple” manipulation of spending or nominal values does not automatically restore higher levels of employment.
Here are less gated versions of the paper.
How sticky are wages in nominal terms?
There is a new paper on this topic, with what seems to be very good data:
This paper examines the relationship between downward nominal wage rigidity and employment outcomes using linked
employer-employee data. Wage rigidity prevents 27.1 percent of counterfactual wage cuts, with a standard deviation of 19.2 percent across establishments. An establishment with the sample-average level of wage rigidity is predicted to have a 3.3 percentage point higher layoff rate, a 7.4 percentage point lower quit rate, and a 2.0 percentage point lower hire rate. Estimating a structural model by indirect inference implies that the cost of a nominal wage cut is 33 percent of an average worker’s annual compensation.
That is by Gabriel Ehrlich and Joshua Montes, recently published in American Economic Journal: Macroeconomics. Here are less gated versions of the paper. I found this of particular interest:
Establishments in the construction supersector display the least wage rigidity, with an average of 8.8 percent of wage cuts prevented. Establishments in the public administration and finance supersectors display the most wage rigidity, with average levels of 39.3 and 41.7 percent of wage cuts prevented, respectively.
Of course employment in the construction sector is highly cyclical, and employment in public administration often is governed by tenure, not exactly what fits best into the standard story.
The Role of Friends in the Opioid Epidemic
Your friends are not always good for you:
The role of friends in the US opioid epidemic is examined. Using data from the National Longitudinal Survey of Adolescent Health (Add Health), adults aged 25-34 and their high school best friends are focused on. An instrumental variable technique is employed to estimate peer effects in opioid misuse. Severe injuries in the previous year are used as an instrument for opioid misuse in order to estimate the causal impact of someone misusing opioids on the probability that their best friends also misuse. The estimated peer effects are significant: Having a best friend with a reported serious injury in the previous year increases the probability of own opioid misuse by around 7 percentage points in a population where 17 percent ever misuses opioids. The effect is driven by individuals without a college degree and those who live in the same county as their best friends.
That is from a new NBER working paper by Effrosyni Adamopoulou, Jeremy Greenwood, Nezih Guner, and Karen Kopecky.