🚀 We’re excited to announce that our paper, “STAN-LOC: Visual Query-based Video Clip Localization for Fetal Ultrasound Sweep Videos,” has been accepted to #MICCAI2024! 🎉

True for other fields too, not just robotics. Most interview pipelines are flawed and aren’t designed to find candidates who genuinely care and have this attention to detail.

Back from #COLM2026 after representing Amazon 🚀 One of my favorite conferences this year—great size for actually meeting people and having real research conversations. Some of the post-training chatter 😅 “SFT causes catastrophic forgetting.” “SFT is still the best. Just lower your LR and add a KL penalty.” “OPSD / OPD? Doesn’t work well. Stay away—just do SFT.” “Bro, just do RL.” Different camps, strong opinions, and surprisingly convincing arguments from all sides 😂 If we crossed paths at COLM, reach out, would love to keep the conversations going 👋

The schedule for our Tokenization workshop today @COLM_conf ! Lots of exciting speakers. Next up: @zamir_ar 🎉

At #COLM2026 this week! 👋 I’ll be at the Amazon Science booth for Meet the Scientist — come say hi and chat about LLM post-training, agentic post-training, multimodal learning, world models, video understanding, or anything AI/ML! 📍 Wed, Oct 7: 12–1 PM 📍 Thu, Oct 8: 1–2 PM

❤️ Learning from Every Heartbeat This #WorldHeartDay, discover our research using AI and fetal echocardiography to develop new approaches for detecting and assessing congenital heart disease before birth. @Ox_wrh 🎥 Explore our research & video vimeo.com/1230199413

This is so fucking cool

Congratulations to @gulvarol and @y_m_asano 🎉! We saw this coming while both were at VGG 😊

Many congratulations @y_m_asano. Very well deserved🥳🥳. Really grateful and honored to have worked with you on so many projects.

I am incredibly honoured to receive the ECVA/ @eccvconf Young Researcher Award. ✨🎉I cannot quite believe it, but what I do know is, that this is all thanks to my amazing collaborators, mentors and students and friends. Thank you! You all put the Fun in Fundamental AI. It’s an adventure to be in this field, in this time.

Cover slide for my talk tomorrow at 9:00am at Malmö Arena, Panorama 2! 🎉 Right after @paschalidoud_1 and before @MattNiessner. ..and yes, I'm one of those folks who shares some cool unpublished results, so it'll be worth coming by ;) @eccvconf

For years, doctors noticed something strange about vaccines in India. Vaccines that worked well in Europe and America worked less well in Indian children. Same vaccine, same dose, weaker protection. Most people accepted it. One scientist in Vellore spent 30 years asking why. Her name is Gagandeep Kang. She trained as a doctor at Christian Medical College in Vellore, in Tamil Nadu, and then did something unusual for a physician. She went into the laboratory and stayed there. Her subject was the gut. Specifically, the guts of children in poor households. She studied what repeated diarrhoeal infection does to a child's intestine over the first years of life. What she found matters enormously. Those repeated infections do not just make a child ill. They damage the lining of the gut, stunt growth, and interfere with how the body responds to oral vaccines. That is why a vaccine developed in a clean environment abroad underperforms in a child who has had 10 infections before the age of 2. The problem was never the vaccine. It was what the child's gut had already been through. Then she went further than diagnosis. Rotavirus is the leading cause of severe diarrhoea in small children and it kills a very large number of them. India had access to imported rotavirus vaccines, but they were expensive and often out of reach for the families who needed them most. Kang led the clinical trials for Rotavac, India's first indigenous rotavirus vaccine, made by an Indian company. She was also central to the work behind a second one, Rotasiil. Both are now approved by the World Health Organisation. Rotavac went into India's national immunisation programme, which means it reaches children whose parents pay nothing for it. In 2019 she became the first Indian woman scientist working in India to be elected a Fellow of the Royal Society, the scientific body founded in London in 1660. It took 359 years. She has served on the World Health Organisation's global advisory committee on vaccine safety, and she now leads global health work at one of the largest health foundations in the world. She built her career on a question most people did not think was worth asking. Not how do we get vaccines to poor children. But why do they not work as well when they get there.

I am glad to present our latest paper, "Adapting MLLMs for Nuanced Video Retrieval" at ECCV 2026 @eccvconf! In this paper, we improve video search for nuanced queries. We define nuance in queries across three axes: (i) temporal: distinguishing between time-sensitive actions such as "folding vs. unfolding paper", (ii) negation: handling simple negators, (iii) multimodal: queries with image/video and text together. We propose a unified model, TARA, based on fine-tuning an MLLM with carefully chosen text-triplets alone. No visual data is used during fine-tuning! TARA achieves strong results on all benchmarks for nuanced retrieval. Visit our Poster: #84 at 4:30PM on the 10th September. Links in the next tweet.

[🔁🚨PhD opening] We have a new opening for a PhD student at our @FunAILab at the University of Technology Nuremberg on combining Vision-Language-Action Models with 3D Geometry. This PhD position is in close collaboration with @HelsingAI with the amazing researcher @TomUnterthiner and will be a unique opportunity to conduct research and have impact. Please see below for more information: jobs.utn.de/jobposting/f75c1… We particularly encourage applications from people of diverse and underrepresented backgrounds. If you have any doubt if you should apply or not, send me a message.

Application deadline is the end of this month‼️🚨

You can just RL a coding model to paint with javascript btw

Hi @duolingo 💚 Any chance you can help a little Duo fan out? My 10-year-old son had a 300+ day streak, but was unexpectedly hospitalized and missed his lessons. He’s home and recovering, but heartbroken about losing his streak. Any chance some Duo magic can restore it? 🥹🔥

Self-Supervised Learning of Structured Dynamics from Videos TL;DR: Structured Dynamics Model decomposes video changes into primary and residual motion. Built on frozen DINOv2 features, it separates camera and object motion with weak supervision and outperforms prior representations on multiple motion tasks. arxiv.org/abs/2607.21576

🇰🇷 Excited to be attending #ICML2026 in Seoul, Korea! 🎉 I’d love to chat about video understanding, self-supervised representation learning, streaming video understanding, post-training VLMs/multimodal LLMs. If you’re working on any of these topics, let’s connect!

"Stay curious and persistent, remain thoughtful and meticulous, and the ideas will come." Very nice article!

SO SO PROUD to see that Gemini Omni Flash is ranked 1st for both Text-to-Video and Image-to-Video on @arena 🥳 Lots of sleepless nights went into making this model a success 💪💪 #gemini #omni @GoogleDeepMind @GoogleAI @Google #ai #genai