Head Google Research. VP Google. Gödel Prize. Kanellakis Award. working on the magic cycle of research; advancing AI as an amplifier of human ingenuity
Google
Joined May 2009
- Tweets1.4K
- Following1K
- Followers5.4K
- Likes3.1K
And we continue asking the boldest questions..
Yossi Matias retweeted
Notebook has just unlocked over 700 textbooks from the world’s leading publishers. Every single one is available to you, completely free of charge! @Gemini_Notebook
Yossi Matias retweeted
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
A long-standing frontier in AI: how can models continuously acquire new knowledge without catastrophic forgetting?
Inspired by how the human brain consolidates memory during rest, our team introduced research on an alternating "wake" and "sleep" lifecycle for LLMs—enabling models to absorb new facts into long-term parameters and self-improve through offline dreaming.
Read more in the blog by by @behrouz_ali and @mirrokni: research.google/blog/the-sle…
Today on the blog, learn about our new instance of a “sleep” phase in continual learning, which enables LLMs to continuously integrate new facts and enhance reasoning while mitigating catastrophic forgetting or static knowledge cutoffs.
More →goo.gle/4iICkyD
ALT Overview of our sleep framework: Input streams from the active phase are stored in short-term memory before entering the sleep cycle, which bifurcates into Consolidation (transferring fragile short-term tokens into expanded long-term parametric memory using low-rank structures and MoE) and Dreaming (autonomous self-improvement via synthetic routing with random experts and RL refinement).
Yossi Matias retweeted
Very excited to share Genie 4, our incredible interactive world model, bringing together real-time physics, reasoning, and audio! Simulations within simulations, makes you wonder just how deep the rabbit hole goes... I have a feeling we will be finding out soon enough...
Tuberculosis claims 1.23M lives every year. At #IDWeek2026, we’re sharing how we are moving from early research benchmarks toward open foundation models that could be tuned by our partners and deployed at the point of care - from clinics across Africa to 4 km underground in gold mines.
👉 goo.gle/4zQZLM1
In a study in @JAMANetworkOpen, @GoogleResearch evaluated how "blind sweep" ultrasonography and on-device AI can help non-sonographers deliver expert-level prenatal assessments—cutting training time to hours. 🧵
Blog: goo.gle/4jrYxlw
Study: jamanetwork.com/journals/jam…
The need is urgent: @WHO estimates 250k+ preventable pregnancy-related deaths yearly, and 2/3 of people globally lack ultrasound access.
Tested with @NorthwesternMed & @jacarandakenya, the model matched expert accuracy with zero local fine-tuning.
Covered also in "Ask a Scientist" episode, watch on Youtube: youtube.com/watch?v=Mk7WJ4XQ…
A new Nature paper by our teams on the Personal Health Agent (PHA)—the foundational multi-agent AI architecture behind Google Health Coach. By orchestrating specialized agents across wearable and biomarker data, we're establishing a benchmark for safe, personalized health AI—and open-sourcing the framework for the community.
Nature paper: nature.com/articles/s41586-0…
An essential aspect of accelerating scientific discovery is grounding AI agents in physical biological reality—moving beyond in-silico hypothesis generation to closed-loop wet-lab experimentation. 🧬🤖
We’re sharing key building blocks for grounding AI agents in biological reality
• VERID: Semantic digital twins for physical robotic safety and lab execution
• Lab-in-the-loop exploration: Autonomous sequence design and self-correction across experimental rounds
• Targeted protein discovery: Uncovering single-digit nanomolar binders for oncology targets like FAP
Explore the research👇
We are bringing agents to the wet-lab bench at Google Research to navigate complex biological landscapes. Learn how we are grounding agents in physical lab robotics to run fast closed-loop molecular design cycles, ultimately discovering single-digit nanomolar protein binders for cancer targets →goo.gle/4B5mMMB
We’re announcing the 17 selected organizations for the Google.org Impact Challenge: AI for Science—receiving $36M in funding + Google.org Accelerator support to tackle critical challenges in climate and health.
Congratulations to all 17 recipient teams! I am excited to see how these interdisciplinary efforts push the boundaries of scientific discovery for societal impact.
Meet the 17 selected organizations and their work: goo.gle/gic-ai-science
Meet the 17 recipients of the Google.org Impact Challenge. These teams are taking on some of science's hardest problems, from disease prevention to mapping the natural world.
Learn how they're unlocking scientific breakthroughs: goo.gle/gic-science
This week at our Research@ flagship event, we announced new updates for our experimental Geospatial Reasoning agent and Earth AI capabilities.
Following promising tests with partners, we are expanding early access to our experimental agent to more partners.
In a new technical report, we demonstrate how combining domain-specific foundation models, spatial tools, and upfront planning enables a Geospatial Reasoning agent to carry out complex spatial analysis and planetary prediction that standalone LLMs can’t.
Highlights from this work👇🧵
--> Real-World Impact with Partners:
Milken Institute: Identifying the highest-impact community resilience hubs for solar energy build out—layering rooftop yield with transit access, heat vulnerability, and local demographics data to compress 6-week siting feasibility studies into 15-min reports to unlock $179 million in clean energy capital.
WHO AFRO & DRC’s INRB: During the recent Ebola outbreak response in the DRC, teams mapped remote mining corridors by cross-referencing satellite forest loss, mining locations, and mobility data—pinpointing 48 exposed settlements and 45,500+ at-risk people in minutes rather than weeks, and modeling transmission risks to plan mobile laboratory deployment.
People in Need: Analyzed historical cyclone paths to identify high-exposure villages in minutes rather than days of manual GIS work, unlocking vital time for anticipatory action.
GiveDirectly: Evaluated population exposure across wind-speed zones during Cyclone Maila and benchmarked tornado damage in Mississippi within a single workspace.
WPP: Connected population dynamics with proprietary data to optimize geospatial marketing while preserving privacy.
Blog announcement: goo.gle/4s13F25
Technical report: arxiv.org/abs/2610.15750
Early access interest form: forms.gle/3kp1t9zq7BqvYxwNA
Exciting milestone from our Google Quantum AI team, with new research published today in Science!
Creating "magic states" has been one of quantum computing's most severe bottlenecks, traditionally threatening to consume up to 95% of a processor's capacity.
In new experimental work on our Willow processor, the team validated "magic state cultivation" on hardware for the first time, achieving:
- 40x error reduction vs. un-cultivated states
- 99.99% logical T-state fidelity
- Successful grafting into a distance-5 surface code
This transforms a 20-year theoretical barrier into an experimental reality, drastically reducing the physical overhead required for useful, fault-tolerant quantum computing.
Congrats to the entire team!
In a crisis, a few seconds of warning can save lives.
Today, we're expanding the Android Earthquake Alerts System from 98 countries to nearly 200 regions! 🌍📱
We’re also improving our latest “Be Ready” alert to give you more information in seconds.
Read more in Android Earthquake blog: blog.google/products-and-pla…
Our crisis resilience efforts also help people prepare for other natural disasters - from flood forecasting to wildfire detection to extreme weather.
When I started on crisis resilience 10 years ago, predicting floods at scale felt impossible.
Today, our global model forecasts river floods up to 7 days ahead for 2B+ people, while FireSat tracks wildfires in near-real-time. Extending warning horizons everywhere.
Our north star: No one should be surprised by a natural disaster. ✨
Google connects people with crisis info >10M times/day.
Learn how we’re working with partners in our new Crisis Resilience White Paper co-authored with Kate Brandt: sustainability.google/report…
