Search

In an online experiment with N = 1893 participants, we collected and analyzed over 18,000 prompts and over 300,000 images to explore how the importance of prompting will change as the capabilities of generative AI models continue to improve. Each participant in our experiment was randomly and blindly assigned to use one of three text-to-image diffusion models: DALL-E 2, its more advanced successor DALL-E 3, or a version of DALL-E 3 with automatic prompt revision. Participants were then asked to write prompts to reproduce a target image as closely as possible in 10 consecutive tries. We find that task performance was higher for participants using DALL-E 3 than for those using DALL-E 2. This performance gap corresponds to a noticeable difference in the similarity of participants' images to their target images, and was caused in equal measure by: (1) the increased technical capabilities of DALL-E 3, and (2) endogenous changes in participants' prompting in response to these increased capabilities. More specifically, despite being blind to the model they were assigned, participants assigned to DALL-E 3 wrote longer prompts that were more semantically similar to each other and contained a greater number of descriptive words. Furthermore, while participants assigned to DALL-E 3 with prompt revision still outperformed those assigned to DALL-E 2, automatic prompt revision reduced the benefits of using DALL-E 3 by 58\%. Taken together, our results suggest that as models continue to progress, people will continue to adapt their prompts to take advantage of new models' capabilities.

Social networks affect the diffusion of information, and thus have the potential to reduce or amplify inequality in access to opportunity. We show empirically that social networks often exhibit a much larger potential for unequal diffusion across groups along paths of length 2 and 3 than expected by our random graph models. We argue that homophily alone cannot not fully explain the extent of unequal diffusion and attribute this mismatch to unequal distribution of cross-group links among the nodes. Based on this insight, we develop a variant of the stochastic block model that incorporates the heterogeneity in cross-group linking. The model provides an unbiased and consistent estimate of assortativity or homophily on paths of length 2 and provide a more accurate estimate along paths of length 3 than existing models. We characterize the null distribution of its log-likelihood ratio test and argue that the goodness of fit test is valid only when the network is dense. Based on our empirical observations and modeling results, we conclude that the impact of any departure from equal distribution of links to source nodes in the diffusion process is not limited to its first order effects as some nodes will have fewer direct links to the sources. More importantly, this unequal distribution will also lead to second order effects as the whole group will have fewer diffusion paths to the sources., Comment: 47 pages

Urban income segregation is a widespread phenomenon that challenges societies across the globe. Classical studies on segregation have largely focused on the geographic distribution of residential neighborhoods rather than on patterns of social behaviors and interactions. In this study, we analyze segregation in economic and social interactions by observing credit card transactions and Twitter mentions among thousands of individuals in three culturally different metropolitan areas. We show that segregated interaction is amplified relative to the expected effects of geographic segregation in terms of both purchase activity and online communication. Furthermore, we find that segregation increases with difference in socio-economic status but is asymmetric for purchase activity, i.e., the amount of interaction from poorer to wealthier neighborhoods is larger than vice versa. Our results provide novel insights into the understanding of behavioral segregation in human interactions with significant socio-political and economic implications.

In this paper, we discuss and analyze our approach to the Fragile Families Challenge. The challenge involved predicting six outcomes for 4,242 children from disadvantaged families from around the United States. The data consisted of over 12,000 features (covariates) about the children and their parents, schools, and overall environments from birth to age 9. Our approach relied primarily on existing data science techniques, including: (1) data preprocessing: elimination of low variance features, imputation of missing data, and construction of composite features; (2) feature selection through univariate Mutual Information and extraction of non-zero LASSO coefficients; (3) three machine learning models: Random Forest, Elastic Net, and Gradient-Boosted Trees; and finally (4) prediction aggregation according to performance. The top-performing submissions produced winning out-of-sample predictions for three outcomes: GPA, grit, and layoff. However, predictions were at most 20% better than a baseline that predicted the mean value of the training data of each outcome.

This study provides the first confirmation that individual employment status can be predicted from standard mobile phone network logs externally validated with household survey data. Individual welfare and households vulnerability to shocks are intimately connected to employment status and professions of household breadwinners. At a societal level unemployment is an important indicator of the performance of an economy. By deriving a broad set of novel mobile phone network indicators reflecting users financial, social and mobility patterns we show how machine learning models can be used to predict 18 categories of profession in a South-Asian developing country. The model predicts individual unemployment status with 70.4 percent accuracy. We further show how unemployment can be aggregated from individual level and mapped geographically at cell tower resolution, providing a promising approach to map labor market economic indicators, and the distribution of economic productivity and vulnerability between censuses, especially in heterogeneous urban areas. The method also provides a promising approach to support data collection on vulnerable populations, which are frequently under-represented in official surveys.

The MapReduce distributed programming framework has become popular, despite evidence that current implementations are inefficient, requiring far more hardware than a traditional relational databases to complete similar tasks. MapReduce jobs are amenable to many traditional database query optimizations (B+Trees for selections, column-store- style techniques for projections, etc), but existing systems do not apply them, substantially because free-form user code obscures the true data operation being performed. For example, a selection in SQL is easily detected, but a selection in a MapReduce program is embedded in Java code along with lots of other program logic. We could ask the programmer to provide explicit hints about the program's data semantics, but one of MapReduce's attractions is precisely that it does not ask the user for such information. This paper covers Manimal, which automatically analyzes MapReduce programs and applies appropriate data- aware optimizations, thereby requiring no additional help at all from the programmer. We show that Manimal successfully detects optimization opportunities across a range of data operations, and that it yields speedups of up to 1,121% on previously-written MapReduce programs., Comment: VLDB2011