🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

I will be in Singapore next week (July 21-25). I am mostly free on Thursday and Friday (24/25). If you are around and want to meet/chat, please let me know!

One more paper in the FB 2020 election series Removing political ads from feeds found no measurable change in knowledge, turnout, polarization, or donations The big lever many campaigns rely on may be more about targeting loyalists than persuasion nber.org/papers/w33818

This video is amazing. Talks about the future of generative video. It's not all 'brainrot' or 'slop'. there's going to be (or maybe already is) a huge potential for a new type of influencer.

I wrote about the use of AI in influence operations/misinformation/propaganda based on reports from OpenAI and Anthropic. Good news: There are small scale trials of using AI but things seem to be under control. I talk about what can be done in the piece gvrkiran.substack.com/p/ai-assisted-…

Interesting methods processing visual content from Instagram dl.acm.org/doi/pdf/10.114…

This is a great piece discussing the strenghts and weakenesses of community moderation on social media (like community notes) arxiv.org/abs/2505.20067

Excited to co-organize the Bridging Human and LLM Annotations for Social Science tutorial at #ic2s2 tomorrow! Join us at 1:30pm CEST to learn how to combine Qualtrics+Prolific, MTurk, & LLMs for automated annotation and valid estimates. Code&materials: github.com/kristinagligor…

This paper presents some nice methods to scalably understand video content. They find a large prevalence of opiod use myths on youtube. Advances in VLMs are making such great analysis possible, which was not possible even 5 years ago. arxiv.org/abs/2506.00308

🚨 Study alert! We've developed a prototype that enables you to create personalized social media feeds using natural language rules and import them into Bluesky. Join our study to try out the tool, give feedback, and earn a $50 gift card! Sign up: docs.google.com/forms/d/e/1FAI…

Agentic AI is supposedly the future, and there are a lot of predictions of massive productivity gains. This paper evaluates does a meta-analysis of 84 studies: only 15 % pair accuracy with human outcomes. arxiv.org/abs/2506.02064

🧵 Academic job market season is almost here! There's so much rarely discussed—nutrition, mental and physical health, uncertainty, and more. I'm sharing my statements, essential blogs, and personal lessons here, with more to come in the upcoming weeks! ⬇️ (1/N)

As AI is being increasingly used everywhere, we need to think of ways to measure and calibrate our dependence on AI (especially for students). I wrote about one way of thinking about it: open.substack.com/pub/gvrkiran/p…

LLMs are really good at working with street view imagery (e.g. to predict road safety, identify businesses, and other factors etc). This paper builds a pipeline for large scale urban analytics using LLMs. arxiv.org/abs/2506.02242

This paper analyzes EU's AI act. The act’s research carve-outs are narrow. If you fine-tune a public model and share code, you’ve likely crossed into “provider” territory. They provide detailed understanding on what researchers should do to prepare. arxiv.org/abs/2506.03218

We (academics) complain a lot about X being a right wing echo chamber these days... (at least I do) This paper shows that Bluesky is an echo chamber in its own ways. Though surprisingly only 13% of the content is political (similar to other social media) arxiv.org/abs/2506.03443

This paper presents a nice methodology to audit automation by AI. By collecting data from 1,500 workers, 104 occupations, and 844 tasks they document mismatched expectations. This paper is a nice take on the thinking beyond AI risk = job loss. arxiv.org/abs/2506.06576

This ACL paper presents a new take on alignment as a solution for AI safety. ‘value alignment’ sounds benign, but their experiments amplify harmful content via cognitive consistency: once a value is endorsed, the model elaborates it with conviction. arxiv.org/abs/2506.06404

This paper shows that Computer Science has increasingly lead work on societal impacts of AI in the past decade. I think it's a healthy sign of maturation for CS Hopefully there will be more interdisciplinary social-science work to avoid echo chambers arxiv.org/abs/2506.08738

One more paper showing that Tiktok's research API (released as part of DSA) still has issues This is the problem with DSA. Companies will find ways to provide useless data & its a huge challenge for independent researchers to audit and ensure compliance arxiv.org/abs/2506.09746