Kai Zhu

Age and gender distortion in online media and large language models - Nature Stereotypes of age-related gender bias are socially distorted, as evidenced by the age gap in the representations of women and men across various media and algorithms, despite no systematic age differences in the workforce.

🧪 Women are represented as younger than men across occupations/social roles in data from Google,Wikipedia,IMDB, Flickr,YouTube and LLMs.The bias is strongest for occupations with high status/earnings.ChatGPT perpetuates this bias when generating and evaluating resumes. www.nature.com/articles/s41...

The New AirPods Can Translate Languages in Your Ears. This Is Profound.

Translation technology is incredible — it’s the #1 AI use case, and I’ve believed that for years.

The demand is huge, and the tools are almost there. With better interactive UI design, the upside for society will be significant.

www.nytimes.com/2025/09/18/t...

This @beijingpalmer.bsky.social post reminded me I have been meaning to update @pewresearch.org's estimate of total newspaper newsroom jobs - which was 30,820 as of 2020

Using BLS OEWS data (w/ same methodology as Pew) the 2024 number is 29,260 - down another 5%

www.pewresearch.org/chart/sotnm-...

Call for papers, 4th UK Workshop on Digital Economics, London 28 November 2025: competitionpolicy.ac.uk/events/4th-w... This is always a great event

How language is hiding the real internet from you Most of the internet is out of your reach, but the barrier isn't just algorithms. In another language, the same platforms turn into whole other worlds.

I wrote an article about linguistic bias and the internet for the BBC, based on a paper @ze.vin, @ethanz.bsky.social, and I wrote comparing four language-specific samples of YouTube. www.bbc.com/future/artic...

Want a good starting point for learning good principles of Dataviz?

I'd highly recommend @andrew.heiss.phd course--Data Visualization with R.

Reading materials, slides, lecture videos, examples, code, etc. are all posted for free on his website.

Law and ethics Post-API Age XML and JSON IP and HTTP Static web pages Archives web pages Dynamic web pages PDFs Wikipedia Government APIs Social APIs Automation AI APIs

Got around to pushing all my @cuboulder.info Web Data Science @jupyter.org notebooks to @github.com

Enjoy! github.com/cuinfoscienc...

There have been a number of recent articles on statistical power in quantitative political science. This is something that I think deserves more attention and discussion. A short thread of the articles I have read. 🧵

Perplexity is using stealth, undeclared crawlers to evade website no-crawl directives Perplexity is repeatedly modifying their user agent and changing IPs and ASNs to hide their crawling activity, in direct conflict with explicit no-crawl preferences expressed by websites.

Perplexity is repeatedly modifying their user agent and changing IPs and ASNs to hide their crawling activity, in direct conflict with explicit no-crawl preferences expressed by websites. blog.cloudflare.com/perplexity-i...

Read the full (Open Access!) paper here: doi.org/10.1287/mnsc...

Thanks to my co-authors, Qiaoni Shi and Shrabastee Banerjee!

Key takeaway: Introducing entry costs can reshapes the ecosystem. Platforms must weigh short-term revenue against the long-term risks of marginalizing small creators, reducing diversity, and harming consumer matching.

The Mismatch: Why lower ratings? We used a fine-tuned BERT model to analyze review text. The results suggest an increase in consumer-book mismatches.

With reduced diversity (a shrinking "long tail"), readers were more likely to receive books misaligned with their preferences.

The Demand Paradox: How did readers (demand side) respond? The promotional effects intensified, but with a paradox.

Books in the paid program received a HIGHER volume of reviews, but LOWER average ratings. Monetization amplified the "Groupon effect."

The "Rich-Get-Richer" Dynamic: Diving deeper, we saw a "rich-get-richer" effect. Popular genres (like Mystery/Thriller) became more dominant, while niche genres (like Poetry/Science) lost market share. The entry cost narrowed the range of cultural products being promoted.

The Diversity Drop: This shift in suppliers directly affected product variety. We measured a significant decline in the diversity of book genres available in the program post-monetization. The marketplace became less varied.

Market Concentration & Author Profiles: This led to a massive 200% increase in market concentration (HHI). Furthermore, the authors who continued to participate post-monetization were generally more established, popular, and experienced with the platform.

Uneven Impact & Concentration: Importantly, the impact was uneven. The cost disproportionately pushed out indie publishers and self-published authors.

While overall participation dropped, the market share of the "Big 5" publishing houses more than doubled (12% to 30%).

The Supply Shock: The impact on the supply side was immediate and dramatic. Introducing the entry cost caused the average number of monthly promotional campaigns to plummet from ~3,000 to ~1,000.

We studied the Goodreads "Giveaways" program, a marketplace for book promotion. It was free for authors/publishers until Jan 2018, when Goodreads introduced a fixed $119 entry cost.

This provided a natural experiment to study monetization in a two-sided market.

📣 Thrilled to announce our new paper, "Monetizing Platforms: An Empirical Analysis of Supply and Demand Responses to Entry Costs in Two-Sided Markets," is now published in Management Science!

When a digital platform starts charging for access, who wins and who loses? 🧵👇

Inspiring PDW on using sensitivity analysis in empirical management research. My contribution is to present the sensemakr package by Cinelli & Hazlett (2020) for observational designs. Thanks a lot to the organizers for putting this fantastic session together. #AOM2025

Where Congress’s Cuts Threaten Access to PBS and NPR The loss of federal funding threatens scores of public TV and radio stations across the United States.

Where the cuts are going to be felt the most.
www.nytimes.com/interactive/...

24/ An excellent recent survey revisits the theoretical literature on herds & cascades It notes that cascades cause poor information aggregation, lead to fragile mass behaviors, and remain central to understanding social learning. Those 1992 papers launched a vast literature 😎

Academia will form these little pockets -- people whose theorizing is outrageous & supported by methods outdated since the 90s -- but once it reaches a critical size those people just review each others papers & grants, form societies, hand out awards etc, like a self-contained parallel society.

its crazy how dominant germany was in science (especially chemistry) in the 19th century, it was basically the international language for scientists, people came from all over the world to train at heidelberg etc, and then....

How museums are using Wikipedia to archive marginalized art As public archival institutions fade and the state rewrites curricula, the next generation of cultural workers is stepping up

Wikipedia may seem like an unlikely site of resistance, but in an era of escalating censorship, disinformation, & erosion of public trust, Wikipedia is a model for collective governance led by next generation of cultural workers, @alexmar.bsky.social reports: prismreports.org/2025/05/12/w...

Wikipedia Contributions in the Wake of ChatGPT How has Wikipedia activity changed for articles with content similar to ChatGPT following its introduction? We estimate the impact using differences-in-differences models, with dissimilar Wikipedia ar...

Just out in WWW last week! 📜Our work on substitution patterns between Wikipedia and ChatGPT. We find *heterogeneous* impacts, where Wiki articles that are similar to ChatGPT outputs see a greater drop in views than dissimilar articles:

arxiv.org/abs/2503.00757

Have you ever asked yourself about the overall extent of TikTok? Here some numbers from "Just Another Hour on TikTok" - Great compliment to @bendavidsteel.bsky.social for this data collection effort!
w/ @miriamschirmer.bsky.social & Derek Ruths
arxiv.org/abs/2504.13279

Climate Terminology Does Not Matter Our new paper finds that swapping out one climate term for another does not meaningfully change people’s stated commitment to fight climate change

Climate Terminology Does Not Matter

Across tens of thousands of participants in two large-scale experiments, we found that labeling climate change in different ways had no effect on their stated willingness to act.
jayvanbavellab.substack.com/p/climate-te...

via @dgoldwert.bsky.social

Wow!

Three scholars at Columbia, Michigan, & Maryland just introduced a measure of the partisan leanings of employers in the U.S.

The data is constructed by linking voter registrations to online worker profiles.

VRscores capture the political affiliations of 21.8M workers across 2.6M employers.