Oriol Vinyals

@husjerry1 @quocleix TIL! Ooops. Is this true in all the tables?

Founder Mode activated from my new office ♾

Thirteen years ago, a quick email exchange with @JeffDean brought me to @Google. After an unforgettable journey from Brain, to DeepMind, and to @GoogleDeepMind, working on seq2seq or distillation, and co-leading huge projects such as AlphaStar, and Gemini, I'm saying goodbye to GDM. The best part? The same person who brought me in is challenging me to take the next leap. Excited to build Discovery Loop (@DiscoLoopAI) alongside @JeffDean, @quocleix, and @Sanjay_Ghemawat Deeply grateful to everyone who made GDM feel like home. Time to try something new and to keep it drastic! ✨♾

@JeffDean @quocleix Fun game indeed with great company. Growing up in Barcelona it was great to see Messi and Yamine together in the field. GGWP!

@migtissera @redtachyon I approve of that use : ) Access to the logits makes learning more efficient but you are still accessing model probabilities through samples. (With lots of assumptions eg sampling method, lacking reasoning traces etc).

Fun fact: @woj_zaremba wanted to completely remove that paper from existence a year or so after. I forgot why though -- Woj? x.com/OfirPress/stat…

This is my third Google I/O in a row. Some things have changed, some (like the amazing colleagues that were there since day 0) haven't. @JeffDean @borgeaud_s

1/ Today at Google I/O, we’re launching Gemini 3.5 Flash ⚡️⚡️⚡️! Our mission was clear: bring frontier-level intelligence with unprecedented speed. 3.5 Flash delivers drastic intelligence (beating 3.1 Pro on almost every benchmark), at Flash speeds. 🧵

2/ What excites me most is 3.5 Flash’s breakthrough performance on complex, multi-step agentic tasks: it excels on Terminal-Bench 2.1 (76.2%), GDPval-AA (1656 Elo) and MCP Atlas (83.6%) 3.5 Flash is built to be the ultra-fast reasoning engine powering your AI agents. The better the model, the better the agent you build on top of it.