Danishjeet Singh

Logo

singhdan [at] iu [dot] edu
Github
Twitter/X
LinkedIn
Google Scholar
My Resume

Website *heavily* inspired by Eugenie Y. Lai

Hi, just call me Danish(daa-nish)

I can be best reached through my email or my LinkedIn.

I’m a first year Computer Science PhD Student at Indiana University, advised by Dr. Mary Jean Amon.

I obtained my undergraduate degree from IU as well and was very fortunate to be advised by Dr. Filippo Menczer at Observatory On Social Media at Indiana Univeristy(OSoMe@IU). Previously I’ve also worked with Dr. David Crandall at the IU Computer Vision Lab.

I’m broadly fascinated by Generative AI, both how people use it and the pitfalls it can create, and I’m committed to digging into this technology at a deeper level to build customizable, explainable systems. From a computational social science perspective, I study the societal harms of generative models like deepfakes and toxic content on social platforms. On the mechanistic side, I probe vision-language models for object hallucination and concept evolution. Ultimately, I want to bridge both worlds: understanding the social issues these tools raise and uncovering their inner workings so we can design safer, more transparent AI.

Publications and Pre-prints

Task-aligned prompting improves zero-shot detection of AI-generated images by Vision-Language Models (2025)
Kachwala, Z., Danishjeet Singh, Yang, D., & Menczer, F.
arxiv preprint
TLDR; Simply prefacing the AI’s response with “Let’s examine the style and the synthesis artifacts” makes vision‑language models 8 %–29 % more accurate at spotting AI-generated pictures compared with the same models used off‑the‑shelf or with generic chain‑of‑thought prompts

Characteristics and prevalence of fake social media profiles with AI-generated faces (2024)
Yang, K.C., Danishjeet Singh & Menczer, F.
Published in the Journal of Online Trust and Safety.
TLDR; A tiny fraction of Twitter profiles use AI‑generated face photos, yet those accounts often disseminate scams and spam; by noting that such fake faces have their eyes in exactly the same place, we could identify them and estimated about 10,000 active accounts of this type.

Recent News [See All Updates]

2025.05 New pre-print alert, Task-aligned prompting improves zero-shot detection of AI-generated images by Vision-Language Models out on arxiv now.

2025.05 I graduated from IU Luddy with a Bachelor of Science in Computer Science, and I’m extremely happy with how much I’ve grown over the past four years, now I’m excited to tackle even bigger challenges ahead.

2025.04 Introducing urloong, comically loong url for your special website!

Projects for creative and serious work!!

Tools & Software that some may find helpful.

Miscellaneous shenanigans basically