Excited to share our latest paper, accepted as a spotlight at the #ICML2024 mechanistic interpretability workshop! We find evidence that LLMs use language-agnostic representations of concepts 🧵↘️

Are you interested in computational social science? Ahead of IC2S2 this year, nature Nature Cities Nature Computational Science and Nature Communications have put together a Focus highlighting some of our favorite recent papers. Check them out! nature.com/collections/di…

Finally out, with Luis Aguiar Jonas Hannane Hannes Ullrich Christian Peukert Returns to Data: Evidence from Web Tracking diw.de/documents/publ… #Webtracking data are essential for consumer profiling that enables targeted advertising & content delivery. 1/6

📣Thrilled that our study on the effectiveness of #LLMs in providing research feedback is now published in NEJM AI! Updated paper: ai.nejm.org/stoken/default…

Love this work from Isabelle Langrock looking at social science in the “post-API” age. She looks at how CSS work has shifted since Facebook closed their Pages API. Important implications as we loose access to Twitter, Reddit’s Pushshift, FB’s crowdtangle and others IC2S2

How can we extend economic complexity ideas to software? Fifteen years ago, we introduced economic complexity metrics using international trade data. This was an important contribution, but trade data was not free of problems. On the one hand, it is not great at capturing

Check our new paper on market power in consumer profiling. Predictive power of behavioral data (web tracking) varies. Large firm can add demographic data, which boosts accuracy. Privacy regulation hits small firms without demographic data harder.

To help explain the weirdness of LLM Tokenization I thought it could be amusing to translate every token to a unique emoji. This is a lot closer to truth - each token is basically its own little hieroglyph and the LLM has to learn (from scratch) what it all means based on

⚠️ New paper alert ⚠️ We study the Digital Markets Act Impact on Google Maps with Michelangelo Rossi (cesifo.org/en/publication… ) A thread 1/n 🧵

the audacity of Nature to say that scientists should slow down, do less, and take more time to think when the pressure to publish in Nature (and other top journals) is why scientists are juggling so much to begin with

It's happening again!! Call for Papers: 2025 Workshop on Platform Analytics platformanalytics.org Davide Proserpio Minkyung Kim

These recent slides from Susan Athey and Guido Imbens at NBER are a great recent review of the most valuable data science methods I'm aware of. They cover tons of ground with lots of pointers. conference.nber.org/confer/2024/SI…

Excited to share our new paper on the design, implementation, and analysis of social media experiments! These experiments let researchers study digital platforms without needing firm collaboration. We hope it's useful to grad students and anyone else who's social media curious!

Happening now! We have a packed room! #PatentPDW #AOM2024

Our new working paper where we analyze a decade long experiment on Facebook where 0.5% of users never see ads: papers.ssrn.com/sol3/papers.cf…

Sharing a new working paper with Anqi Zhao & Peng Ding Peng Ding, titled "Factorial Difference-in-Differences." arxiv.org/pdf/2407.11937… 🧵 Comments and suggestions are welcome!

Re-reading Lundberg et al. which I always assign in the first week of intro methods. It's so polite, but its core point amounts to the most savage of critiques: studies should be very precise about the question they are trying to answer, but that usually doesn't happen.

Very happy to note that my paper (w/ #XingyiLi, #YitingDeng & Bert De Reyck) on the impact of expert ratings is now forthcoming at #ManagementScience! We are very grateful to the review team, led by DE #EricAnderson, which was tough but fair, providing constructive advice

In case you missed it ravibapna and I had a wide ranging conversation about Enterprise AI Strategy and our book #Thrive with Michael Krigsman #CXOTALK CXOTALK youtu.be/cmoc8CTR_DU?si…