Today we discuss knowledge profiling, a behavioral framework that measures both encoding and recall, and use it to examine the underlying bottlenecks of factuality in frontier LLMs. Read more →https://goo.gle/4wt4pxE
About us
From conducting fundamental research to influencing product development, our research teams have the opportunity to impact technology used by billions of people every day. We aspire to make discoveries that impact everyone, and sharing our research and tools to fuel progress in the field is fundamental to our approach.
- Website
-
https://research.google/
External link for Google Research
- Industry
- Technology, Information and Internet
- Company size
- 1,001-5,000 employees
Updates
-
Today, we’re announcing the 2026 Google Carbon Removal and Superpollutant Elimination R&D Awards. Google is providing funding to support high-impact, open R&D across twelve priority areas and one open call — from ocean alkalinity enhancement to atmospheric chemistry. If you are an academic researcher, non-profit, or innovator working to advance these critical mitigation strategies, we want to partner with you. Applications are open until September 25, 2026. Learn more and apply: https://goo.gle/3TRhGCB
-
-
We’ve advanced AMIE, our research medical AI system, to support real-time video consultations. Our latest study demonstrates expert-level performance in a randomized, controlled trial with 300 simulated consultations. Read the blog to learn more: goo.gle/3RWLpcP https://lnkd.in/gVkakqfr
Advancing AMIE Towards Expert-Level Video Consultations
https://www.youtube.com/
-
Google Research reposted this
Today, in a paper published in Nature, we show that the WeatherNext AI model by Google DeepMind and Google Research teams achieved state-of-the-art accuracy in predicting a cyclone's track, intensity, and wind structure. We’ve also open sourced the code and model weights today, making them freely available for all. On average, the model extends accurate forecasting by an additional day - it can predict three days ahead with the accuracy previous models achieved with only two. This scale of improvement corresponds roughly to a decade’s worth of meteorological progress historically. ✨ Bridging Global and Local Scales: Traditional physics-based Numerical Weather Prediction (NWP) models have historically struggled to efficiently handle both massive global atmospheric currents (which steer a cyclone's track) and highly localized processes around its core (which contribute to drive intensity). WeatherNext Cyclones bridges this gap as a single AI model trained end-to-end on global atmospheric analyses and historical storm databases. ✨ Uncertainty Modeling with FGNs: Using Functional Generative Networks (FGNs), the system can now generate 1,000-member ensembles using less than a TPU-minute per sample, empowering forecasters to evaluate devastating tail-risks like rapid intensification events. ✨ Surprising Efficiency: While traditional specialized intensity models require ultra-high resolutions (such as 2x2km pixels), WeatherNext Cyclones only needs a coarser 28x28km input resolution proving that extreme high resolution is not strictly required for state-of-the-art intensity forecasting. ✨ Real-World Impact & Open Source: During the 2025 Atlantic hurricane season, WeatherNext helped the National Hurricane Center make a historic forecast for Hurricane Melissa five days in advance. To accelerate the field, the model weights and code for WeatherNext Cyclones and WeatherNext 2 are now open source-including a lightweight model that runs on a single TPU in a public Colab notebook. WeatherNext showcases the potential of AI to help build more resilient communities, providing local forecasters and emergency responders worldwide with the tools they need to proactively protect lives and infrastructure. Read our paper in Nature: https://lnkd.in/g9vGRVcT Full announcement from today: https://lnkd.in/g_FCjtHV Explore the data on Weather Lab: https://lnkd.in/gEqf_7Cv Github repo: https://lnkd.in/ghfYvKdq
-
-
Today at 5:00pm, you have one more chance to win bragging rights! Join Sebastian Bodenstein at the #DLI2026 Google booth for the final session of "Tokens, Transformers and Trivia: The AI Research Foundations Quiz" at Deep Learning Indaba. Test your knowledge of modern language models and discover a free learning path to become research-ready.
Ready to test your AI knowledge? Today at 1:30pm, join Sebastian Bodenstein and AIMS at Deep Learning Indaba Google booth for "Tokens, Transformers and Trivia: The AI Research Foundations Quiz"! Drawn from the Google DeepMind AI Research Foundations course, this fast-paced quiz will test your intuition on modern language models, from tokenization to fine-tuning. Prizes and bragging rights included! #DLI2026
-
-
Did you miss the first WeatherNext demo? Today at 2:30pm, join Ilan Price at the Deep Learning Indaba Google booth to see Google DeepMind's AI-based weather and tropical cyclone forecasting models in action. Learn about the underlying technology and see real-time performance from last year's cyclone season. #DLI2026
How can AI improve weather forecasting? Today at 2:30pm, join Ilan Price at the Deep Learning Indaba Google booth for a demo of WeatherNext, showcasing Google DeepMind's AI-based weather and tropical cyclone forecasting models. We'll discuss the underlying model and review case-study storms from last year's cyclone season to see its real-time performance. #DLI2026
-
-
Second chance to catch this session! If you missed the earlier Q&A, join Alex Goldin today at 1:30pm at the Deep Learning Indaba Google booth to explore the sociotechnical implications of AGI. Discover recent work from the Post-AGI Research Team at Google DeepMind and share your thoughts on research priorities. #DLI2026
What are your research priorities as we prepare for AGI? Today at 2:30pm, join Alex Goldin at the Deep Learning Indaba 2026 Google booth for a Q&A and discussion exploring the sociotechnical implications of AGI for the world. Alex will be sharing recent work from the Post-AGI Research Team at Google DeepMind. Don't miss this insightful conversation! #DLI2026
-
-
How can AI improve weather forecasting? Today at 2:30pm, join Ilan Price at the Deep Learning Indaba Google booth for a demo of WeatherNext, showcasing Google DeepMind's AI-based weather and tropical cyclone forecasting models. We'll discuss the underlying model and review case-study storms from last year's cyclone season to see its real-time performance. #DLI2026
-
-
Ready to test your AI knowledge? Today at 1:30pm, join Sebastian Bodenstein and AIMS at Deep Learning Indaba Google booth for "Tokens, Transformers and Trivia: The AI Research Foundations Quiz"! Drawn from the Google DeepMind AI Research Foundations course, this fast-paced quiz will test your intuition on modern language models, from tokenization to fine-tuning. Prizes and bragging rights included! #DLI2026
-
-
Applications for GARA are closing this week on Aug 7. Submit your proposals now for Google academic research awards, a global open call for proposals that support groundbreaking foundational and applied research in computing and technology. This cycle, Google is interested in two quantum computing topics: • Early Fault-Tolerant Algorithms & Applications • Quantum Computing Security This is an opportunity for up to $100K in research funding and to connect with industry researchers. Learn more and apply at goo.gle/gara.