Past: Research @GoogleDeepMind (Gemini 3) | @iiit_hyderabad
Montréal
Joined October 2021
- Tweets68
- Following526
- Followers209
- Likes2.7K
Darshan Singh retweeted
Great work, important to benchmark on the right things! Happy to see that VELOCITI, our CVPR 2025 work survives! 🎉
Try it out, short 10 second videos that stump many models.
katha-ai.github.io/projects/…
Darshan Singh retweeted
Excited to be at ICML'26 🇰🇷 from July 6 onward!
I'll be presenting papers on Reasoning, RL and Diffusion Language Models.
Would love to chat about RL for agents, text-optimization, self-improvement, and diffusion LMs.
See you in Seoul 🚀
Paper threads and details below 👇
Darshan Singh retweeted
Check out this new work by Darshan and gang! Really enjoyed working on non-western, non-english domains with such an amazing team. At CVPR this year!
Frontier models have become excellent at understanding videos. But what happens when we test them outside the comfort zone of Western, English-centric data? In our #CVPR2026 (Highlight) work, we pushed these models to their limits to see if they can function effectively in diverse global contexts. The results? They are struggling. Work done with @NagraniArsha @skawshik11 @Harman26Singh @dinesh_tewari1 @0xtob @CordeliaSchmid Anelia Angelova @shachi_dave (1/7)
Darshan Singh retweeted
The Project Aria team is excited to be part of the Third Joint Egocentric Vision (EgoVis) Workshop at #CVPR2026! 👓✨
Don’t miss the latest Project Aria updates as we dive into the future of egocentric perception. Check out the full program of invited speakers below!
📅 June 3, 2026
📍 Colorado Convention Center | Room 704/706
🔗 egovis.github.io/cvpr26/
⚡️@liuziwei7 , @mapo1 , @doughty_hazel
Darshan Singh retweeted
Spending billions to train the "best" base model? You might be optimizing the wrong thing! 🎯
We show that controlling sharpness during mid-training leads to over 35% less forgetting after fine-tuning / quantization... even when the base model itself gets worse.
🧵 Takeaways for pretraining:
- Use SAM (Sharpness-Aware-Minimization) in the final steps (~10%)
- Try much higher learning rates (yes, even ~10× larger)
1/9
Darshan Singh retweeted
Can LLMs Self-Verify? Much better than you'd expect.
LLMs are increasingly used as parallel reasoners, sampling many solutions at once.
Choosing the right answer is the real bottleneck.
We show that pairwise self-verification is a powerful primitive.
Introducing V1, a framework that unifies generation and self-verification:
💡 Pairwise self-verification beats pointwise scoring, improving test-time scaling
💡 V1-Infer: Efficient tournament-style ranking that improves self-verification
💡 V1-PairRL: RL training where generation and verification co-evolve for developing better self-verifiers
🧵👇
Darshan Singh retweeted
Our paper was desk rejected @NeurIPSConf! Even before the main deadline! "Non-academic title and abstract" 🙈
Thankfully, @SIGGRAPHAsia was around the corner and a perfect fit for our work on improving robustness of multi-subject multi-attribute layout-guided T2I models!
🧵1/9
Darshan Singh retweeted
Researchers consider themselves very successful if they win one test-of-time award (and one is more than enough). Ross @inkynumbers has been winning them nonstop over the past year: CVPR 2024, ICCV 2025, and now NeurIPS 2025, because winning just one was too easy for him!
Having known him for many years (first as a climbing partner and then as a colleague), I can’t say I’m surprised. When he sets his mind to something, he perfects it, whether it is making the best vision model, climbing a 5.12d, or continuing the sally-up sally-down push-up challenge until the rest of the team gives up.
And to all his collaborators who only worked with him remotely and didn’t get to see him in person every day: you missed out. He is fun to work with but he is even more fun in person. I'm attaching the proof below. I have some true gem videos of his goofy side that I won’t share (saving them for when I need to blackmail him), but here is a photo of Ross pretending to be a lizard under our office sun lamp.
Congratulations to Ross and all his co-authors.
#NeurIPS2025
Darshan Singh retweeted
The NeurIPS Test of Time Award recognizes papers that have made lasting contributions to machine learning. For 2025 the award goes to…(buff.ly/W4ohNLA) #NeurIPS2025 #NeurIPSanDiego"
ALT Kaiming He
Darshan Singh retweeted
Late life update 🚀 I started my PhD at @UCBerkeley after an incredible time at @GoogleDeepMind. It was exciting to work on Gemini over the past couple of years.
These days I am interested in reasoning/improving RL, agents, and diffusion language models. Looking forward to contributing to open science. Also thrilled to be back in the Bay Area.
Grateful to mentors, collaborators, and folks who supported me, @partha_p_t @PengchuanZ @nitish_gup @trevorcohn @xiangrenNLP @divy93t @ManishGuptaMG1, Parag Singla, friends, and family.
Excited to be at @NeurIPSConf #NeurIPS2025 this week. Looking forward to meeting folks. Feel free to DM if you'd like to chat!
Darshan Singh retweeted
Really proud of what we have achieved with Gemini 3 🚀!
The Gemini MM team has worked relentlessly across image 🖼️ and video 🎥 from pre-training to post-training to simply deliver the best multimodal in the world 👏!
Looking forward to what you will build🫡!
Darshan Singh retweeted
This is Gemini 3: our most intelligent model that helps you learn, build and plan anything.
It comes with state-of-the-art reasoning capabilities, world-leading multimodal understanding, and enables new agentic coding experiences. 🧵
Darshan Singh retweeted
Happy that @gaur_manu and @thought2vec have this opportunity for our work on revisiting self-retrieval for fine-grained image captioning!
Thread 🧵: xitter.cf/gaur_manu/status/18534…
arXiv: arxiv.org/abs/2409.03025
Project page: katha-ai.github.io/projects/…
We at TMLR are proud to announce that selected papers will now be eligible for an opportunity to present at the joint NeurIPS/ICML/ICLR Journal-to-Conference (J2C) Track:
medium.com/@TmlrOrg/tmlr-joi…
Darshan Singh retweeted
The goal of any good theory or new knowledge is to reduce uncertainty or entropy of a field. Given how many papers published each year related to machine/artificial intelligence and how many different opinions about this subject, I hope our new book is a denoiser, not a diffuser.
Darshan Singh retweeted
Excited to share I have joined CMU’s Robotics Institute as a Master’s student! The past few months were tough with visa uncertainties, but I’m happy to be finally here 😇
I look forward to working on native visual reasoning + other cool RL and VLM stuff 🚀
If you’re around campus, let’s hang out!
Darshan Singh retweeted
Still can’t believe it. Our work on uncovering plagiarism in AI generated research received the outstanding paper award at ACL!!
Effort led by, and envisioned by the amazing @tarungupta360!
