Today, I'm releasing videos in the series on how I use AI in my life. The series is a collection of videos on how i use AI tools like ChatGPT, Claude and Gemini for research, ideation, writing, image/video/audio generation, data collection, gaming, etc. https://t.co/RP5y7qdxeV

this is a great dataset of Youtube videos which trended on the platform from over 100 countries spanning 3 years https://t.co/pngjqLcJTz

This is a nice paper discussing the issue with most social media studies -- we always over sample the small fraction of active users. The paper discusses how biased we could be, the effects of inference on such data and ways to mitigate this. https://t.co/rGRabMATQu

Want to be an intern at Microsoft Research in the Computational Social Science group in NYC (Jake Hofman, David Rothschild, Dan Goldstein) Follow this link and do your thing! Deadline approaching soonish! https://t.co/wKFBQmhLzt

I wrote about what the end of ad supported web means. Thoughts/comments welcome. https://t.co/V7yfRYQazk

AI agents vs humans at work: agents are ~88% faster and 90-96% cheaper, but worse quality and prone to data fakery. So even though they're not great, they can be deployed in low stakes, easily programmable tasks, before they come for the big jobs. https://t.co/7SYLYNQEdD

this paper makes a nice argument about the increasing difficulty of detecting AI generated text. what does AI generated text mean when humans are learning and changing our patterns according to the AI and vice versa? https://t.co/XcGV3Vq1Bs

Though real world use of LLMs for mental health is quite low (~5% of the queries, according to the recent OpenAI paper), this paper finds through a survey that around a quarter of the participants use it for mental health related uses. https://t.co/thJ7yo4N1K

This paper shows that around 9% of mainstream news articles are at least partially AI generated. As I wrote in my recent substack post, defining what AI generated content is will become harder as we start using these tools more often. https://t.co/E8CYLJH4HA

This paper specifically looks at AI non adopters (around 2/3rd of US adults) and suggests that their needs are very different. They make a compelling argument that looking at usage logs and not focusing on non adopters will lead to very biased outcomes. https://t.co/Yav7XKBgop

We might have seen this chart showing that the web is now over 50% "AI generated". I've been seeing a lot of commentary on how the web is dead. In this piece, I think about why that may not be the case. https://t.co/4JF3CF96t1

This paper studies that the Twitter API was used in over 30k studies! Though we mostly see the loss of the API as a loss of important data, an important side effect of the loss is that the community stopped being hooked to this free and very biased data. https://t.co/SoAxYfemjI

This paper on "Generative Propaganda" powerfully argues that our collective obsession with "deepfakes" is distracting us from the real threat. AI's efficiency in supercharging the creation of persuasive (and often obvious) content is a bigger problem. https://t.co/unXZJrgJ76

Tawfiq @TawfiqAmmari presenting our paper on how students use ChatGPT at ASIST 2025. We obtained real world logs from Rutgers students to show unique patterns in usage. More results in the paper: https://t.co/BtYNOCRvY0

A thought-provoking paper by philosophers on "AI Wellbeing." They argue that combining major theories of mind & wellbeing, existing AI systems already have a welfare level My priors that this is a future sci-fi problem were revised after reading the paper https://t.co/6xIcZKIV20

Another cool paper from .@RutgersCommInfo's @gvrkiran tells us about his #community-driven #fact-checking system on #WhatsApp #asist25 #assist2025

I wrote about why blaming AI data centers for your high electricity bill isn't the full story. They're a convenient scapegoat, but a big problem is the ancient, slow, grid we're plugging them into. https://t.co/gP0K2rgsa6

This paper finds AI models are overly agreeable, which makes users less willing to repair real-life interpersonal conflicts Worrying that people *prefer* this AI. They rate sycophantic AI higher and trust it. creates a feedback loop, rewarding validation https://t.co/36F9yXmSnJ

This study does a "blind test" asking people to rate content quality if generated by Humans and AI and finds significant pro-human attribution bias. Content written by humans and labeled as AI gets rated very low! https://t.co/TUWhU0HS4b

This paper shows that over 1/4th of the top1k websites restrict AI crawlers. There are key distinctions: e.g. only 4.1% of right-leaning outlets block OpenAI crawlers. What happens when most of your news accessible to the crawlers is right leaning? https://t.co/3e5o8WkiSa

Last week I wrote about AI companions and why they might have a lot of positive use cases. I think the next step for AI companions is porn. and it's going to be a big business. I tried to think about what that means. https://t.co/4OnalsALDK