Introducing LSM-2, our newest foundation model for wearable sensor data. LSM-2 uses Adaptive & Inherited Masking, a novel self-supervised framework, to learn from incomplete data & achieve strong performance without requiring explicit imputation. More → goo.gle/4kP6Ncc

Congratulations to the authors of "Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction", which won an #ICML2025 Outstanding Paper Award! arxiv.org/abs/2504.15266

First wildfire images from FireSat are here! FireSat spotted an Oregon fire undetected by existing systems, showing its ability to catch small fires in their earliest stages — an important milestone in our partnership w/ Earth Fire Alliance & Muon Space. More →goo.gle/44RFzeY

Today on the blog we discuss our recent research on privacy-preserving synthetic data in #FederatedLearning to improve both small and large language models in #Gboard production, which benefits user typing experiences in practice. Read all about it →goo.gle/3ILPnQz

Google is proud to be a Diamond Sponsor of the 2025 ACL 2025 on natural language processing and understanding, held this week in Vienna, Austria! Attending #ACL2025? Learn more about our work, including papers, workshops, and booth activities at goo.gle/46llnEX

Attending #ACL2025? Stop by the Google booth (9D) to learn about groundbreaking research from computational approaches to natural language understanding, attend a demo, or chat with researchers. Find the schedule of booth activities at goo.gle/46llnEX

Let your wearable data "speak" for itself! Introducing SensorLM, a family of sensor-language foundation models trained on ~60 million hours of data, enabling robust wearable data understanding with natural language. → goo.gle/4lSLwQi

Stop by the #ACL2025 Google booth today at 10AM when Omer Goldman & Reut Tsarfaty will teach you about ECLeKTic, a new benchmark designed to evaluate the ability of LLMs to transfer knowledge across different languages. More at goo.gle/3Y5TqvZ & arxiv.org/abs/2502.21228

Drop by the #ACL2025 Google booth today at 3:30 when Naman Goyal & Heidi Zhang will present interactive webapps & games created end-to-end via vibe coding on Gemini Canvas, the creative platform where you can unleash your imagination. gemini.google/overview/canva…

It is the final day of the #ACL2025 Expo! In attendance? Don't miss your chance to stop by the Google booth to meet some researchers and learn about our latest research.

Announcing AlphaEarth Foundations, part of our Earth AI collection of geospatial models that tackle critical planetary needs: ☁️ Weather 🌊 Floods 🔥 Wildfires 🏙️ Urban planning 🏥 Public health 🚗 Urban mobility See the blog from Yossi Matias & Chris Phillips:

Congratulations to the authors of "AfriMed-QA: A Pan-African, Multi-Specialty, Medical Question-Answering Benchmark Dataset", recipient of the Best Social Impact Award at #ACL2025! aclanthology.org/2025.acl-long.…

Our new regression language model approach enables LLMs to predict industrial system performance directly from raw text data. Learn how we applied it to Borg, Google’s large-scale compute infrastructure for cluster management, & copy the code. → goo.gle/3JaS7Ha

Presenting MLE-STAR, a novel research focused ML engineering agent that integrates web search and targeted code block refinement that could help foster innovation and streamline ML model development. Learn more at goo.gle/4fmXvmK

Introducing DeepPolisher — in collaboration with UC Santa Cruz Genomics Institute, we developed this new open-source method to improve genome assembly accuracy. It reduces indel errors by 70% and total assembly errors by 50%. Learn more: goo.gle/3Hn8PCG

Type 2 diabetes affects hundreds of millions globally. Our research explores a suite of ML models that show the potential of using wearable data & routine blood tests to effectively predict insulin resistance, a major precursor to type 2 diabetes. More: goo.gle/4oDX9MS

Today on the blog we feature a new active learning method for curating high-quality data that reduces training data requirements for fine-tuning LLMs by orders of magnitude. Read all about it → goo.gle/3H2mrDv