We hope these findings can initiate an honest dialogue within the community concerning the ethical and reliable utilization of AI-generated text. Joint work with UMD Department of Computer Science students: Vinu Sankar Sadasivan Aounon Kumar Sriram B Wenxiao Wang

📢 Paper update: despite recent claims, AI-text detectors are still quite unreliable! In our updated draft, we show "recursive paraphrasing" can break new detectors based on watermarking and text retrieval. Paper: arxiv.org/abs/2303.11156 Details below: x.com/feizisoheil/st…

Vision Transformers used to be much better than CNNs in handling input masking since they could just drop the masked tokens In our #ICCV2023 work, we close this gap by developing improved input masking for CNNs: arxiv.org/abs/2211.14646 Joint work with Sriram B ; details 👇

AI giants like Google use image watermarking to combat deepfakes We now show that: (i) existing methods are quite vulnerable against diffusion, adversarial & spoofing attacks (ii) imperceptible watermarks can never become reliable Paper: arxiv.org/abs/2310.00076 A 🧵

Deep networks can be incredibly oversensitive to even small input perturbations, so we might expect that they'd at least be sensitive to large semantic changes. In our #NeurIPS2023 Spotlight paper, we show that general vision models (CNNs, VITs) have significant “blind spots”!🧵

Similar to mode connectivity in weight space, deep networks have extensive connections in image space too. Check out Gaurang Sriramanan 's thread for more details, or come talk to us in today's evening poster session at stall 916!

NEVER USING OFX again! OFX : I sent money to my family in India for some urgent family reasons on January 7th. Thread:

Presenting ArtSavant 🎨, a practical tool for interpretably detecting and articulating artistic style infringements. 🔗 arxiv.org/abs/2404.08030 Joint work w/ UMD, GW Law, and NSF Trails Institute

AI System Achieves Silver Medal-level score in IMO The International Mathematical Olympiad (IMO) is the oldest, largest & most prestigious competition for young mathematicians. Every year, countries send their top young mathematicians to take a 6 problem test spanning two days.