Introducing a new roadmap for building highly reproducible AI benchmarks. In collaboration with researchers at RIT, we investigated the trade-off between the number of items and human raters per item — the forest vs. the tree. Our research suggests that the common standard of 3–5 raters is often insufficient for capturing human nuance. We found that by optimizing the rater distribution based on specific metrics, practitioners can achieve statistically reliable results and better account for natural disagreement in subjective tasks like safety and toxicity detection. We have open-sourced our simulation tool to help the community design more efficient and reliable evaluations. Learn more: https://goo.gle/4bQC8u5
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
-
This #NationalAILiteracyDay, we’re proud to see Google Research projects like AI Quests — developed in collaboration with the Stanford Accelerator for Learning — helping students apply AI literacy to solve real-world problems!
This National #AILiteracyDay, we’re helping you move from "in the loop" to "in the lead" with resources designed to help educators, learners, families, and more navigate the world of AI safely and responsibly. 💻 Professional AI learning to all 6 million U.S. educators through our partnership with ISTE+ASCD 📝 Gemini Faculty Fundamentals to master your academic workflows 🎮 Immersive, gamified learning for your students with AI Quests Explore these free resources and more today to help build the next generation of AI-literate leaders: ai.google/literacy
-
How can AI help bridge global healthcare gaps? In the MedGemma Impact Challenge, launched in collaboration with Kaggle, 850+ teams showcased the potential. Today, we’re announcing the winners! Building upon our open-weight models, these developers are tackling a diverse range of critical healthcare challenges: 🏆 1st Place: EpiCast – designed for detecting disease outbreaks in West Africa. 🥈 2nd Place: Sunny – designed for privacy-first skin cancer screening. 🥉 3rd Place: FieldScreen AI – designed for on-device tuberculosis screening. 🏅 4th Place: Tracer – designed for the prevention of medical errors. ✨Special technology winners: ClinicDx, UniRad3s, BridgeDx, CaseTwin, BigTB6 ✨Honorable mentions: Dual Path ICU, Sentinel, Enso Atlas, CAP CDSS Congratulations to all the winners! Learn more about their innovations →goo.gle/47oQG1l
-
-
Introducing Vibe Coding XR, a new rapid prototyping workflow that empowers Gemini Canvas w/ the XR Blocks framework to turn user prompts into interactive, physics-aware WebXR applications, allowing creators to quickly test intelligent spatial experiences →https://goo.gle/4bw0VTY
-
Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4takpUc
-
-
Mapping the modern world: We introduce S2Vec, a self-supervised framework, part of Google Earth AI, that transforms complex geospatial data into general-purpose embeddings for predicting population density, carbon emissions, and urban development at scale. Check out the blog: goo.gle/41rLzKa
-
-
Today we are expanding the Google Quantum AI research program to include neutral atom quantum computing. This is a strategic move to sit alongside our core superconducting qubit work. By investing in both modalities, we can tap into their complementary strengths. Superconducting qubits offer the fast cycle times necessary for deep circuits, while neutral atom arrays provide a flexible connectivity graph for efficient error correction. Both paths accelerate our mission to build large-scale, error-corrected quantum computing to solve unsolvable problems, an engineering challenge we are excited to tackle. To lead this new chapter, we welcome Dr. Adam Kaufman to the team. Based in the AMO hub of Boulder, Colorado, he will lead a team embedded within the research ecosystem of CU Boulder and NIST. Read the full update from Hartmut Neven on the Google Keyword blog: https://goo.gle/40TycSV
-
Discover the latest quantum computing research at APS Global Physics Summit 2026, March 16–19. From optimized algorithms to superconducting qubits and error characterization, there’s a lot to explore → goo.gle/46UJFoI
-
-
Today on the blog, read all about how a large-scale evaluation of a mammography AI system across multiple NHS screening services in the UK demonstrates its potential to enhance cancer detection accuracy and reduce workload within complex double-reading workflows. More →https://goo.gle/47831qu
-
Today we announce a new era of healthcare innovation at The Check Up, highlighting how Google Research is transforming patient care and scientific discovery. Our latest advancements focus on four key areas: - Clinical Collaboration: Research with @ImperialCollege and the NHS shows our AI system helped radiologists identify 25% of interval cancers that are typically missed during screenings. - Agentic AI: We are testing AMIE, a multi-agent system, in clinical settings to support history-taking and real-time clinical reasoning. - Foundational Models: Our MedGemma open-weight models have been downloaded over 3 million times, empowering developers to build localized health solutions. - Public Health: Using Google Earth AI, researchers are now creating high-resolution estimates of vaccination coverage to help public health teams pinpoint local gaps. We remain committed to the highest standards of accuracy and responsibility, collaborating with @GoogleDeepMind and global health partners to ensure these tools are safe and helpful for everyone. Learn more: goo.gle/4sL6WBS
-