Large language models reduce public knowledge sharing on online Q&A platforms.

Large language models (LLMs) are a potential substitute for human-generated data and knowledge resources. This substitution, however, can present a significant problem for the training data needed to develop future models if it leads to a reduction of human-generated content. In this work, we document a reduction in activity on Stack Overflow coinciding with the release of ChatGPT, a popular LLM. To test whether this reduction in activity is specific to the introduction of this LLM, we use counterfactuals involving similar human-generated knowledge resources that should not be affected by the introduction of ChatGPT to such extent. Within 6 months of ChatGPT's release, activity on Stack Overflow decreased by 25% relative to its Russian and Chinese counterparts, where access to ChatGPT is limited, and to similar forums for mathematics, where ChatGPT is less capable. We interpret this estimate as a lower bound of the true impact of ChatGPT on Stack Overflow. The decline is larger for posts related to the most widely used programming languages. We find no significant change in post quality, measured by peer feedback, and observe similar decreases in content creation by more and less experienced users alike. Thus, LLMs are not only displacing duplicate, low-quality, or beginner-level content. Our findings suggest that the rapid adoption of LLMs reduces the production of public data needed to train them, with significant consequences.

A large-scale empirical investigation of specialization in criminal career.

We use a comprehensive longitudinal dataset on criminal acts over 6 years in a European country to study specialization in criminal careers. We present a method to cluster crime categories by their relative co-occurrence within criminal careers, deriving a natural, data-based taxonomy of criminal specialization. Defining specialists as active criminals who stay within one category of offending behavior, we study their socio-demographic attributes, geographic range, and positions in their collaboration networks relative to their generalist counterparts. Compared to generalists, specialists tend to be older, are more likely to be women, operate within a smaller geographic range, and collaborate in smaller, more tightly-knit local networks. We observe that specialists are more intensely embedded in criminal networks, suggesting a potential source of self-reinforcing dynamics in criminal careers.

Digital traces of brain drain: developers during the Russian invasion of Ukraine.

The Russian invasion of Ukraine has caused large scale destruction, significant loss of life, and the displacement of millions of people. Besides those fleeing direct conflict in Ukraine, many individuals in Russia are also thought to have moved to third countries. In particular the exodus of skilled human capital, sometimes called brain drain, out of Russia may have a significant effect on the course of the war and the Russian economy in the long run. Yet quantifying brain drain, especially during crisis situations is generally difficult. This hinders our ability to understand its drivers and to anticipate its consequences. To address this gap, I draw on and extend a large scale dataset of the locations of highly active software developers collected in February 2021, one year before the invasion. Revisiting those developers that had been located in Russia in 2021, I confirm an ongoing exodus of developers from Russia in snapshots taken in June and November 2022. By November 11.1% of Russian developers list a new country, compared with 2.8% of developers from comparable countries in the region but not directly involved in the conflict. 13.2% of Russian developers have obscured their location (vs. 2.4% in the comparison set). Developers leaving Russia were significantly more active and central in the collaboration network than those who remain. This suggests that many of the most important developers have already left Russia. In some receiving countries the number of arrivals is significant: I estimate an increase in the number of local software developers of 42% in Armenia, 60% in Cyprus and 94% in Georgia. Supplementary Information: The online version contains supplementary material available at 10.1140/epjds/s13688-023-00389-3.

No Country for Asylum Seekers? How Short-Term Exposure to Refugees Influences Attitudes and Voting Behavior in Hungary.

How does exposure to refugees influence political behavior? We present evidence from Hungary, a country with widespread anti-immigration attitudes, that short term exposure during the 2015 refugee crisis predicts anti-refugee voting and sentiment. We code exposure to refugees at the settlement level using reports from state media, an independent online news site, and an online social media aggregator. Settlements through which refugees traveled showed significantly higher anti-refugee voting in a national referendum in 2016. The effect decreases sharply with distance. Using a difference-in-differences model, we find that the far-right opposition gained, while the governing right-wing party lost votes in these settlements in subsequent parliamentary elections. This suggests incumbents are punished by voters skeptical of immigration regardless of their policy position. Survey data supports this finding of a competition among right-wing parties, as individuals in exposed settlements are more fearful of immigrants and support restrictive policies only if they identify as right-wing.

Evolving collaboration, dependencies, and use in the Rust Open Source Software ecosystem.

Open Source Software (OSS) is widely spread in industry, research, and government. OSS represents an effective development model because it harnesses the decentralized efforts of many developers in a way that scales. As OSS developers work independently on interdependent modules, they create a larger cohesive whole in the form of an ecosystem, leaving traces of their contributions and collaborations. Data harvested from these traces enable the study of large-scale decentralized collaborative work. We present curated data on the activity of tens of thousands of developers in the Rust ecosystem and the evolving dependencies between their libraries. The data covers eight years of developer contributions to Rust libraries and can be used to reconstruct the ecosystem's development history, such as growing developer collaboration networks or dependency networks. These are complemented by data on downloads and popularity, tracking dynamics of use, visibility, and success over time. Altogether the data give a comprehensive view of several dimensions of the ecosystem.

Inequality is rising where social network segregation interacts with urban topology.

Social networks amplify inequalities by fundamental mechanisms of social tie formation such as homophily and triadic closure. These forces sharpen social segregation, which is reflected in fragmented social network structure. Geographical impediments such as distance and physical or administrative boundaries also reinforce social segregation. Yet, less is known about the joint relationships between social network structure, urban geography, and inequality. In this paper we analyze an online social network and find that the fragmentation of social networks is significantly higher in towns in which residential neighborhoods are divided by physical barriers such as rivers and railroads. Towns in which neighborhoods are relatively distant from the center of town and amenities are spatially concentrated are also more socially segregated. Using a two-stage model, we show that these urban geography features have significant relationships with income inequality via social network fragmentation. In other words, the geographic features of a place can compound economic inequalities via social networks.

The Effects of Gender Signals and Performance in Online Product Reviews.

This work quantifies the effects of signaling gender through gender specific user names, on the success of reviews written on the popular amazon.com shopping platform. Highly rated reviews play an important role in e-commerce since they are prominently displayed next to products. Differences in reviews, perceived-consciously or unconsciously-with respect to gender signals, can lead to crucial biases in determining what content and perspectives are represented among top reviews. To investigate this, we extract signals of author gender from user names to select reviews where the author's likely gender can be inferred. Using reviews authored by these gender-signaling authors, we train a deep learning classifier to quantify the gendered writing style (i.e., gendered performance) of reviews written by authors who do not send clear gender signals via their user name. We contrast the effects of gender signaling and performance on the review helpfulness ratings using matching experiments. This is aimed at understanding if an advantage is to be gained by (not) signaling one's gender when posting reviews. While we find no general trend that gendered signals or performances influence overall review success, we find strong context-specific effects. For example, reviews in product categories such as Electronics or Computers are perceived as less helpful when authors signal that they are likely woman, but are received as more helpful in categories such as Beauty or Clothing. In addition to these interesting findings, we believe this general chain of tools could be deployed across various social media platforms.

A network approach to cartel detection in public auction markets.

Competing firms can increase profits by setting prices collectively, imposing significant costs on consumers. Such groups of firms are known as cartels and because this behavior is illegal, their operations are secretive and difficult to detect. Cartels feel a significant internal obstacle: members feel short-run incentives to cheat. Here we present a network-based framework to detect potential cartels in bidding markets based on the idea that the chance a group of firms can overcome this obstacle and sustain cooperation depends on the patterns of its interactions. We create a network of firms based on their co-bidding behavior, detect interacting groups, and measure their cohesion and exclusivity, two group-level features of their collective behavior. Applied to a market for school milk, our method detects a known cartel and calculates that it has high cohesion and exclusivity. In a comprehensive set of nearly 150,000 public contracts awarded by the Republic of Georgia from 2011 to 2016, detected groups with high cohesion and exclusivity are significantly more likely to display traditional markers of cartel behavior. We replicate this relationship between group topology and the emergence of cooperation in a simulation model. Our method presents a scalable, unsupervised method to find groups of firms in bidding markets ideally positioned to form lasting cartels.

Social capital predicts corruption risk in towns.

Corruption is a social plague: gains accrue to small groups, while its costs are borne by everyone. Significant variation in its level between and within countries suggests a relationship between social structure and the prevalence of corruption, yet, large-scale empirical studies thereof have been missing due to lack of data. In this paper, we relate the structural characteristics of social capital of settlements with corruption in their local governments. Using datasets from Hungary, we quantify corruption risk by suppressed competition and lack of transparency in the settlement's awarded public contracts. We characterize social capital using social network data from a popular online platform. Controlling for social, economic and political factors, we find that settlements with fragmented social networks, indicating an excess of bonding social capital has higher corruption risk, and settlements with more diverse external connectivity, suggesting a surplus of bridging social capital is less exposed to corruption. We interpret fragmentation as fostering in-group favouritism and conformity, which increase corruption, while diversity facilitates impartiality in public life and stifles corruption.