Amazing opportunity to build SOTA multicultural+multimodal AI!

not missing the chance to get roasted by AI on a weekend

I wish I enjoyed anything as much as o3 enjoys making nice little tables

How can agents trained to reach (temporally) nearby goals generalize to attain distant goals? Come to our #ICLR2025 poster tomorrow to discuss 𝘩𝘰𝘳𝘪𝘻𝘰𝘯 𝘨𝘦𝘯𝘦𝘳𝘢𝘭𝘪𝘻𝘢𝘵𝘪𝘰𝘯! 🗓️Saturday, April 26, 2025 🕙10:00 – 12:30 AM 📍Hall 3 + Hall 2B #637

ICLR Spotlighted! ✨ Canfer Akbulut presenting our research tomorrow at #ICLR2025 on evaluating #AI's understanding of historical images using our new Century dataset.

Spent an amazing afternoon with second graders in a bay area public school, talking about AI + creating stories, videos, lessons, and images with Google Gemini App! My heart is full - the curiosity and creativity of students to explore the utility of these models was unbelievable

Some years ago, I got trapped in a Massive Trough of Imposter Syndrome. It took more than a year to dig myself out of it, but the following framework really helped me. It feels a bit vulnerable to share, but I hope it might help a few others too! A short thread 🧵🙂

Public love for my boy Logan Kilpatrick who does 100x more behind the scenes (than you see here on X) to make Gemini work for developers. I'm finding myself leaning on him more than ever to make things happen!

On my waymo yesterday , I resolved a sticky bank issue on the phone - proud of the effective time use instead of driving. Today, I slept. Maybe AGI would let me sleep more😴

Honoured to judge (pun intended) the #AgentX competition with UC Berkeley RDI. Dawn Song keeps lowering the barrier to AI-agent development - this MOOC is another great example. Bring on those entries! 🚀

I'm thrilled that our key innovation, MatFormers (🪆🪆🪆), now offers many-in-one flexibility for Gemma 3n. This allows us to: a. embed a 2B model within the 4B model, eliminating the need for hosting two separate models for different use cases; b. extract multiple accurate

hype for Claude Plays Pokemon w/Claude 4 starting now! (for fun, Gemini Plays Pokemon is also re-starting at the same time to sync the streams - but keep in mind that the streams are not directly comparable due to different agentic setups!!!)

Congratulations, Aditya Grover ! Making diffusion models go brrr should be added to this list too :D

Introducing Matryoshka Teaching Assistant aka matTA framework! This allows learning a "Matformed TA-student" pair with distillation from a much more relatable TA, and provides elasticity/adaptivity/flexibility of Matryoshka models. Key benefits: a. more accurate, servable

Good to bear in mind that most people you see tweeting about AI papers do not know anything about research or science or technology or actually about AI for that matter

If geeking out on architecture, efficiency, and adaptability in foundation models hypes you up -- go work with the best boy Prateek Jain at Google DeepMind India. He’s an absolute legend to work with. Only con : his dad jokes are 😐 -- but Bangalore weather and food more than

woke up to 3 am calls from friends asking about my family in Ahmedabad, frantically called home and finally was breathing normal when they picked up, so many didn't get that feeling -- my heart goes out to the precious lives lost, the families grieving, and those impacted on the

What a week for AI w/ two amazing start-up launches (happen to be from people I adore) 1/ Scouts by Yutori makes my life easy by giving me an actionable agent that works hard for me to find very interesting papers, news --> some papers it found for me are wild! 2/ Yupp comes