Xavier(Xavi) Amatriain
@xamat
Followers
23K
Following
7K
Media
174
Statuses
1K
AI Product @Google. Cofounder @CuraiHQ. Former @LinkedIn @Quora and @Netflix. Catalan in the Valley. Runner & Ironman.
Silicon Valley/Barcelona
Joined October 2007
For those drawing conclusions on the differences between model responses, remember that these models are stochastic in their responses. The comparison between them has no value unless done over many responses by each model to the same prompt. As an example, find two very
GPT-5, Claude, Kimi, and Gemini: "I can travel back in time to any time before 1500 and change only one thing, what is the single thing you would change, nothing obvious."
0
0
3
Rosalia us the 🐐. Lux is an incredible and very complex album. And, it is still breaking records! 🙌
On November 7, Rosalía’s 'LUX' became the most-streamed album in a single day in Spotify history by a Spanish-speaking female artist.
0
0
3
🗣️Just used @GoogleAIStudio Build + the Live API to create a fully multilingual app for patient intake! As the patient has the conversation with Gemini (in any language!), the data is added to the intake form using function calling. Users can even dynamically switch languages:
13
29
305
Opal, our no-code AI app builder, is now available in 160 countries! See how people are using it to easily create AI applications: https://t.co/jSrOMOcHJQ
blog.google
We’re bringing Opal to 160 countries.
0
0
2
You still hear some people claiming that LLMs "don't reason" or can only "regurgitate content from the internet" 🙄 . Meanwhile, top mathematicians in the world like Terence Tao and Javier Gomez Serrano using Alpha Evolve write: "large language model-guided evolutionary search
arxiv.org
AlphaEvolve is a generic evolutionary coding agent that combines the generative capabilities of LLMs with automated evaluation in an iterative evolutionary framework that proposes, tests, and...
1
2
7
One of your top requests is here: Gemini Deep Research now connects directly to your Gmail, Drive, Docs, and Chat. Available for all users on desktop now (mobile soon) Two common ways people are using it: * Market analysis combining live web trends with your team's internal
68
88
823
Welcome to the Age of Inference! Ironwood TPUs are now GA 🎉 https://t.co/N4x9Pseve0
cloud.google.com
Google Cloud’s compute portfolio now includes Ironwood TPUs and Axion-based N4A VMs and C4A bare metal.
0
0
4
Also really cool to see one of the leads from my teams co-authoring the paper!
0
0
0
DataRater: can you automatically+continuously learn which examples will help the model the most? Fun working on this w/ awesome set of co-authors: @luisa_zintgraf, @dancalian, @greg_far, @iurii_kemae, @matteohessel, @shar_jeremy, @junh_oh, András György, Tom Schaul, @hado, &
Excited to share our new paper, "DataRater: Meta-Learned Dataset Curation"! We explore a fundamental question: How can we *automatically* learn which data is most valuable for training foundation models? Paper: https://t.co/N2ozU2RXWb to appear @NeurIPSConf Thread 👇
10
36
349
Very excited about this (almost literal) moonshot!
Our TPUs are headed to space! Inspired by our history of moonshots, from quantum computing to autonomous driving, Project Suncatcher is exploring how we could one day build scalable ML compute systems in space, harnessing more of the sun’s power (which emits more power than 100
2
0
7
Amazing work, and happy to see several team members included as authors here!
Google Earth AI, our collection of geospatial AI models and datasets, is expanding globally and adding new capabilities. That includes Geospatial Reasoning, powered by Gemini, which automatically connects different Earth AI models - like weather forecasts, population maps +
0
0
2
Today, we’re announcing a major breakthrough that marks a significant step forward in the world of quantum computing. For the first time in history, our teams at @GoogleQuantumAI demonstrated that a quantum computer can successfully run a verifiable algorithm, 13,000x faster than
469
2K
15K
Today, we're introducing the new vibe coding experience in @GoogleAIStudio!🔥 We want to make it as easy as possible for you to build AI apps :) Short demo of what's new:
69
163
1K
I was honored to deliver the keynote at the ACM #RecSys conference ( https://t.co/LLnnS4RDiB) in Prague last month! My talk, "Recommending in the Age of AI," was a personal look at the evolution of the field and where we're heading next. We've come a long way from the early days
0
0
3
Awesome to see Veo 3.1 top the LMArena video leaderboards by a large distance with big improvements over Veo 3.0 for text-to-video (+30) and image-to-video (+70)! 🔥Huge congrats to the team! Try it for yourself in https://t.co/QgTpxTL8DQ and the @GeminiApp
🚨🎬 Big news from Video Arena! @GoogleDeepMind’s latest Veo 3.1 now ranks #1 in both Text-to-Video and Image-to-Video leaderboards. 🏆 This is a +30-point leap from Veo 3.0 → 3.1, making it the first model to break 1400 in Video Arena history! Huge congrats to the
51
108
1K
Very exciting announcement today! You can now enjoy the incredible performance of TPUs with vLLM with your framework of choice. PyTorch and JAX are both fully supported in this new unified backend! 🎉
Announcing the completely reimagined vLLM TPU! In collaboration with @Google, we've launched a new high-performance TPU backend unifying @PyTorch and JAX under a single lowering path for amazing performance and flexibility. 🚀 What's New? - JAX + Pytorch: Run PyTorch models on
0
0
6
Very cool!
Witnessing the magic of "grokking" in action with the new JAX native Tunix library! The port of Grokking (based on implementation initially in MLX) demonstrates how transformers suddenly "understand" modular arithmetic after initial memorization—jumping from 11% to 99% validation
1
0
2
Fantastic to see Genie 3, our state-of-the-art world model, featured in @TIME's 2025 Best Inventions. From a single image or text prompt to an entire playable world, it’s the future of AI and entertainment. So proud of @jparkerholder @shlomifruchter & the team - huge congrats!
We’re proud to announce that Genie 3 has been named one of @TIME’s Best Inventions of 2025. Genie 3 is our groundbreaking world model capable of generating interactive, playable environments from text or image prompts. Find out more → https://t.co/bv1gZaWYtd
39
89
865
The new Computer-Use model from @GoogleDeepMind has been crushing it in trusted benchmarks, We benchmarked the model against the new models from @AnthropicAI as well as @OpenAI on Online Mind2Web & Web Voyager. Gemini crushes them all in accuracy, speed, and cost (by a lot).
35
77
715
Welcome Gemini Enterprise! The new front door for AI in the workplace! https://t.co/Sw3q5zP4uH
0
1
4