Statistical topic modeling is widely used in political science to study text. Researchers examine documents of varying lengths, from tweets to speeches. There is ongoing debate on how document length affects the interpretability of topic models. We investigate the effects of aggregating short documents into larger ones based on natural units that partition the corpus. In our study, we analyze one million tweets by U.S. state legislators from April 2016 to September 2020. We find that for documents aggregated at the account level, topics are more associated with individual states than when using individual tweets. This finding is replicated with Wikipedia pages aggregated by birth cities, showing how document definitions can impact topic modeling results.

Vision Language Models (VLMs) are rapidly advancing in their capability to answer information-seeking questions. As these models are widely deployed in consumer applications, they could lead to new privacy risks due to emergent abilities to identify people in photos, geolocate images, etc. As we demonstrate, somewhat surprisingly, current open-source and proprietary VLMs are very capable image geolocators, making widespread geolocation with VLMs an immediate privacy risk, rather than merely a theoretical future concern. As a first step to address this challenge, we develop a new benchmark, GPTGeoChat, to test the ability of VLMs to moderate geolocation dialogues with users. We collect a set of 1,000 image geolocation conversations between in-house annotators and GPT-4v, which are annotated with the granularity of location information revealed at each turn. Using this new dataset, we evaluate the ability of various VLMs to moderate GPT-4v geolocation conversations by determining when too much location information has been revealed. We find that custom fine-tuned models perform on par with prompted API-based models when identifying leaked location information at the country or city level; however, fine-tuning on supervised data appears to be needed to accurately moderate finer granularities, such as the name of a restaurant or building.

Transformer networks have revolutionized NLP representation learning since they were introduced. Though a great effort has been made to explain the representation in transformers, it is widely recognized that our understanding is not sufficient. One important reason is that there lack enough visualization tools for detailed analysis. In this paper, we propose to use dictionary learning to open up these "black boxes" as linear superpositions of transformer factors. Through visualization, we demonstrate the hierarchical semantic structures captured by the transformer factors, e.g., word-level polysemy disambiguation, sentence-level pattern formation, and long-range dependency. While some of these patterns confirm the conventional prior linguistic knowledge, the rest are relatively unexpected, which may provide new insights. We hope this visualization tool can bring further knowledge and a better understanding of how transformer networks work. The code is available at https://github.com/zeyuyun1/TransformerVis

In this paper, we argue that AI ethics must move beyond the concepts of race-based representation and bias, and towards those that probe the deeper relations that impact how these systems are designed, developed, and deployed. Many recent discussions on ethical considerations of bias in AI systems have centered on racial bias. We contend that antiblackness in AI requires more of an examination of the ontological space that provides a foundation for the design, development, and deployment of AI systems. We examine what this contention means from the perspective of the sociocultural context in which AI systems are designed, developed, and deployed and focus on intersections with anti-Black racism (antiblackness). To bring these multiple perspectives together and show an example of antiblackness in the face of attempts at de-biasing, we discuss results from auditing an existing open-source semantic network (ConceptNet). We use this discussion to further contextualize antiblackness in design, development, and deployment of AI systems and suggest questions one may ask when attempting to combat antiblackness in AI systems.

Fox News correspondent Claudia Cowan joins'Your World' with the details from San Francisco This is a rush transcript from "Your World with Neil Cavuto" June 8, 2021. This copy may not be in its final form and may be updated. NEIL CAVUTO, FOX NEWS ANCHOR: How about some good news to kick off things, like herd immunity happening in a lot of parts of this country, including in San Francisco, where close to eight out of 10 residents older than 12 years old have already had at least one vaccination shot? It reads similarly in other cities, like Philadelphia, 67.4 percent have been vaccinated, in Denver, close to 70 percent, in San Diego, north of 65 percent, and, in New York City, more than 52 percent. And this is "Your World." And FOX on top of vaccinations that are surprisingly robust across a country that is rapidly leading the world in finally putting a spike in this horrific, horrific disease. Now, the implications of all of this are being weighed in the medical community, as well as the political community, as to how much longer term this means we get to, well, herd immunity, if we even need to get to that, technically, at the rate we're going. Let's go to Claudia Cowan following all of this in San Francisco -- Claudia. The City by the Bay is on the cusp of herd immunity, which means that the coronavirus is having trouble finding new hosts. The city is reporting that nearly 80 percent of teens and adults have been vaccinated with at least one dose against COVID-19, while 68 percent are fully vaccinated. The number of new cases is the lowest since the city shut down in March of 2020. And no one has died of COVID in over a month. San Francisco pushed people to get the shot while infections hospitalizations and death rates were low. Officials believe that made a world of difference. While there is some debate over what exactly constitutes herd immunity, one expert says the numbers here are among the best in the country. MONICA GANDHI, UNIVERSITY OF CALIFORNIA, SAN FRANCISCO: And there are places, like in the Bay Area, that are up to 76, 77 percent. So, we are doing great in terms of high vaccination rates, high immunity, low cases, low hospitalizations, low deaths, low test positivity rate.

Robots have been around, in some form or other, since the ancient world. Early legends of automata existed in Greek and Roman legends and basic mechanical'robots' were designed and built in China and Greece. Our modern concept of robots wouldn't appear until the Industrial Revolution with the notion of the android (humanoid robot) coming into existence in 20th Century film and sci-fi literature. Since then many engineers have worked tirelessly to improve and, in some cases, redefine robotics. These 15 are just some of them. The following is far from exhaustive and is in no particular order. Their contribution to robotics: Joseph Endgelberger is widely credited for the birth of the robotics industry.

Rufus Thompson, left, is accused of kidnapping and murdering Danny Diaz-Delgado near Trenton, N.J. last month. A New Jersey man has been arrested and accused of kidnapping and murdering a man trying to buy a PlayStation video game console that was advertised online, according to multiple reports. Rufus Thompson, 29, was arrested Saturday morning in Trenton. He is charged with murder, felony murder, robbery, kidnapping and weapons offenses in the death of 20-year-old Danny Diaz-Delgado. The Trentonian reported that Diaz-Delgado's body was found March 24 near the banks of Assunpink Creek in Hamilton Township.

In 1961, the world's first industrial robot clocked in at a General Motors plant in Trenton, New Jersey. The 4,000-pound mechanical arm, named Unimate, was designed to weld cars and lift big pieces of metal. The robot was a huge success--even landing a spot on the Tonight Show with Johnny Carson. Get the latest from Kellogg Insight delivered to your inbox. Almost sixty years later, the emergence of artificial intelligence (AI) has seen machines leap from physical to mental labor.

In 1961, the world's first industrial robot clocked in at a General Motors plant in Trenton, New Jersey. The 4,000-pound mechanical arm, named Unimate, was designed to weld cars and lift big pieces of metal. The robot was a huge success--even landing a spot on the Tonight Show with Johnny Carson. Almost sixty years later, the emergence of artificial intelligence (AI) has seen machines leap from physical to mental labor. As computers step into roles that involve reasoning, a new wave of industries ranging from medicine to finance stands to benefit from--or get left behind by--AI.

It's long been clear that urbanization and automated technologies are shaping society, but it hasn't been obvious how the two forces affect each other. A new study from MIT's Media Lab posits that the smaller the city, the greater the impact it faces from automation. The finding, they say, could encourage legislators to pay special attention to workers in smaller cities and offer them support services. Other researchers have attempted to measure the effect of technology on employment in cities, but the Media Lab authors, who have identified which jobs and skills tend to be more prevalent in smaller cities and larger ones, claim to be the first to explain why different U.S. cities are more susceptible (or resilient) to technological unemployment. They say that bigger cities have a disproportionately large number of jobs for people who do cognitive and analytical tasks, such as software developers and financial analysts--occupations that are less likely to be disrupted by automation.