google ai research projects

Publications

Explore our research on some of the most complex and interesting challenges in AI.

Latest research news

Discover our latest AI breakthroughs and updates from the lab

View all posts

FermiNet: Quantum physics and chemistry from first principles

Using deep learning to solve fundamental problems in computational quantum chemistry and explore how matter interacts with light

AI achieves silver-medal standard solving International Mathematical Olympiad problems

Breakthrough models AlphaProof and AlphaGeometry 2 solve advanced reasoning problems in mathematics

Google DeepMind at ICML 2024

Exploring AGI, the challenges of scaling and the future of multimodal generative AI

Explore a selection of our recent research on some of the most complex and interesting challenges in AI.

• Date 21 Aug 24 21 August 2024 Title The Vizier Gaussian Process Bandit Algorithm Authors Xingyou Song, Qiuyi Zhang, Chansoo Lee, Emily Fertig, Tzu-Kuo Huang, Lior Belenki, Greg Kochanski, Setareh Ariafar, Srinivas Vasudevan, Sagi Perel, Daniel Golovin Venue arXiv
• Date 13 Aug 24 13 August 2024 Title The Probabilities Also Matter: A More Faithful Metric for Faithfulness of Free-Text Explanations in Large Language Models Authors Noah Y. Siegel, Oana-Maria Camburu, Nicolas Heess, Maria Perez-Ortiz Venue ACL 2024
• Date 7 Aug 24 7 August 2024 Title Achieving Human Level Competitive Robot Table Tennis Authors David B. D'Ambrosio, Saminda Abeyruwan, Laura Graesser, Atil Iscen, Heni Ben Amor, Alex Bewley, Barney J. Reed, Krista Reymann, Leila Takayama, Yuval Tassa, Krzysztof Choromanski, Erwin Coumans, Deepali Jain, Navdeep Jaitly, Natasha Jaques, Satoshi Kataoka, Yuheng Kuang, Nevena Lazic, Reza Mahjourian, Sherry Moore, Kenneth Oslund, Anish Shankar, Vikas Sindhwani, Vincent Vanhoucke, Grace Vesom, Peng Xu, and Pannag R. Sanketi Venue arXiv
• Date 31 Jul 24 31 July 2024 Title Pre-trained Gaussian processes for Bayesian optimization Authors Zi Wang, George E. Dahl, Kevin Swersky, Chansoo Lee, Zachary Nado, Justin Gilmer, Jasper Snoek, Zoubin Ghahramani Venue JMLR
• Date 21 Jul 24 21 July 2024 Title Levels of AGI for Operationalizing Progress on the Path to AGI Authors Meredith Ringel Morris, Jascha Sohl-dickstein, Noah Fiedel, Tris Warkentin, Allan Dafoe, Aleksandra Faust, Clement Farbaret, Shane Legg Venue ICML 2024
• Date 16 Jul 24 16 July 2024 Title Language models, like humans, show content effects on reasoning tasks Authors Andrew K Lampinen*, Ishita Dasgupta*, Stephanie C Y Chan, Hannah R Sheahan, Antonia Creswell, Dharshan Kumaran, James L McClelland, Felix Hill (*equal contribution) Venue PNAS Nexus
• Date 16 Jul 24 16 July 2024 Title Neural Climate Data Compression Authors Piotr Mirowski, David Warde-Farley, Mihaela Rosca , Matthew Koichi Grimes , Yana Hasson, Hyunjik Kim, Mélanie Rey, Simon Osindero, Suman Ravuri and Shakir Mohamed Venue arXiv
• Date 17 Jun 24 17 June 2024 Title Bayes' Rays: uncertainty quantification for neural radiance fields Authors Lily Goli, Cody Reading, Silvia Sellan, Alec Jacobson, Andrea Tagliasacchi Venue CVPR 2024
• Date 17 Jun 24 17 June 2024 Title Mirasol3B: A Multimodal Autoregressive Model for Time-Aligned and Contextual Modalities Authors AJ Piergiovanni, Isaac Noble, Dahun Kim, Michael S. Ryoo, Victor Gomes, Anelia Angelova Venue CVPR 2024
• Date 17 Jun 24 17 June 2024 Title Neural Fields as Distributions: Signal Processing Beyond Euclidean Space Authors Daniel Rebain, Soroosh Yazdani, Kwang Moo Yi, Andrea Tagliasacchi Venue CVPR 2024
• Date 7 Jun 24 7 June 2024 Title Don't trust your eyes: on the (un)reliability of feature visualizations Authors Robert Geirhos, Roland S. Zimmermann*, Blair Bilodeau*, Wieland Brendel*, Been Kim Venue ICML 2024
• Date 5 Jun 24 5 June 2024 Title A Robot Walks into a Bar: Can Language Models Serve as Creativity Support Tools for Comedy? An Evaluation of LLMs' Humour Alignment with Comedians Authors Piotr Mirowski, Juliette Love, Kory Mathewson, Shakir Mohamed Venue FAccT 2024
• Date 28 May 24 28 May 2024 Title An Introduction to Universal Artificial Intelligence Authors Marcus Hutter, David Quarel, Elliot Catt Venue Chapman and Hall/CRC
• Date 8 May 24 8 May 2024 Title Super-Exponential Regret for UCT, AlphaGo and Variants Authors Laurent Orseau, Remi Munos Venue arXiv
• Date 7 May 24 7 May 2024 Title π2vec: Policy Representations with SuccessorFeatures Authors Gianluca Scarpellini*, Ksenia Konyushkova, Claudio Fantacci, Tom Paine, Yutian Chen, Misha Denil Venue ICLR 2024
• Date 7 May 24 7 May 2024 Title Learning 3D Particle-based Simulators from RGB-D Videos Authors Will Whitney, Tatiana López, Tobias Pfaff, Yulia Rubanova, Thomas Kipf, Kimberly Stachenfeld, Kelsey Allen Venue ICLR 2024
• Date 7 May 24 7 May 2024 Title Adaptive Hashing: Faster Hash Functions Perhaps with Fewer Collisions Authors Gábor Melis Venue European Lisp Symposium 2024
• Date 7 May 24 7 May 2024 Title CORRELATED NOISE PROVABLY BEATS INDEPENDENT NOISE FOR DIFFERENTIALLY PRIVATE LEARNING Authors Christopher A. Choquette-Choo, Krishnamurthy Dvijotham, Krishna Pillutla, Arun Ganesh, Thomas Steinke, Abhradeep Thakurta Venue ICLR 2024
• Date 7 May 24 7 May 2024 Title Privacy Amplification by Sampling for the Matrix Mechanism. Authors Christopher A. Choquette-Choo, Arun Ganesh, Thomas Steinke, Abhradeep Thakurta Venue ICLR 2024
• Date 7 May 24 7 May 2024 Title Teach LLMs to Phish: Stealing Private Information from Language Models Authors Ashwinee Panda, Christopher A. Choquette-Choo, Zhengming Zhang, Yaoqing Yang, Prateek Mittal Venue ICLR 2024

Explore our other teams and product areas

LABS.GOOGLE

Experiment with the future of ai, visual artists ✕ imagen.

MUSICFX RADIO

Better than ever

AI in Google products

Revisit familiar tools, now enhanced with ai features.

AI Overviews

Gen AI in Chrome

Help Me Script

Magic Editor

Magic Compose

Learn About

Project IDX

Emerging ai tools, try our experiments around how we create, develop, and learn with ai.

Enjoy our suite of games

Say What You See

Musical Canvas

Poem Postcards

Lab sessions, experimental ai collaborations with innovators to shape technology to meet their own unique needs..

TEACHER X LLMs

JOURNALIST ✕ LLMs

RAPPER ✕ LLM

COMPOSER ✕ GEN AI

STUDENTS ✕ SIGN LANGUAGE RECOGNITION

Green Light

Using google ai to reduce traffic emissions, green light optimizes traffic lights to reduce vehicle emissions in cities, helping mitigate climate change and improving urban mobility.

Road transportation is responsible for a significant amount of global and urban greenhouse gas emissions. It is especially problematic at city intersections where pollution can be 29 times higher than on open roads.  At intersections, half of these emissions come from traffic accelerating after stopping. While some amount of stop-and-go traffic is unavoidable, part of it is preventable through the optimization of traffic light timing configurations. To improve traffic light timing, cities need to either install costly hardware or run manual vehicle counts; both of these solutions are expensive and don’t provide all the necessary information.

Green Light uses AI and Google Maps driving trends, with one of the strongest understandings of global road networks, to model traffic patterns and build intelligent recommendations for city traffic engineers to optimize traffic flow. Early numbers indicate a potential for up to 30% reduction in stops and 10% reduction in greenhouse gas emissions (1). By optimizing each intersection, and coordinating between adjacent intersections, we can create waves of green lights and help cities further reduce stop-and-go traffic. Green Light is now live in over 70 intersections in 13 cities, 4 continents, from Haifa, Israel to Bangalore, India to Hamburg, Germany – and in these intersections we are able to save fuel and lower emissions for up to 30M car rides monthly. 

Green Light reflects Google Research's commitment to use AI to address climate change and improve millions of lives in cities around the world.

How it works

1. understanding the intersection.

Building on our decades-long effort to map cities across the world, we can infer existing traffic light parameters including: cycle length, transition time, green split (i.e. right-of-way time and order), coordination and sensor operation (actuation).

2. Measuring traffic trends 

We create a model to understand how traffic flows through the intersection. This helps us understand typical traffic patterns including patterns of starting and stopping, average wait times at a traffic light, coordination between adjacent intersections (or lack thereof), and how traffic light plans change throughout the day.

3. Developing recommendations for the city

Using AI, we identify possible adjustments to traffic light timing. We share these adjustments as actionable recommendations with the city. The city’s traffic engineers review the recommendations, approve them, and they can easily implement them in as little as 5 minutes, using the city's existing policies and tools.

4. Analyzing impact 

We measure how many stops we’ve saved for drivers, and its impact on traffic patterns. We then use industry standards models to calculate the climate impact of these changes. We share this with the partner city and continue monitoring for any future needed changes.

User friendly recommendation interface for partner cities

The Green Light dashboard provides city-specific actionable recommendations, showing supporting trends for each recommendation, with the option to accept or reject the suggestion. After a recommendation has been implemented, the dashboard shows an impact analysis report.

Why bring Green Light to your city?

A simple, high-impact way for cities to go Green

No additional hardware purchase, installation or maintenance required 

Automatic coverage, monitoring and optimization of intersections    

Trusted source of truth (based on Google Maps driving trends)

Clear actionable recommendations and impact reports

User friendly interface

If you are a city representative or traffic engineer and are interested in joining the waiting list, please complete this form .

Green Light has become an essential component of Kolkata Traffic Police. It serves several valuable purposes which contribute to safer, more efficient, and organized traffic flow and has helped us to reduce gridlock at busy intersections. Since November 2022, we have implemented suggestions at 13 intersections. The outcome is excellent as per the feedback from commuters and traffic personnel

Vineet Kumar Goyal Commissioner of Police, Kolkata, India

Green Light identified opportunities where we previously had no visibility and directed engineers to where there were potential benefits in changing signal timings. This  provided valuable insights for our city with 2,400 traffic signals. Both the Green Light and Transport for Greater Manchester teams brought expertise and ideas to the table to improve journeys and reduce emissions. 

David Atkin Analysis and Reporting Manager, Transport for Greater Manchester, England

The system collects, analyzes and presents information that otherwise would require lots of time and effort to manually process. Changes can bring immediate benefits. Green Light helps engineers in their decision making process and enables quick insights.

Anat Gilad Director of the Traffic Control and Technology Management Division, Haifa, Israel

Green Light in the news

How Google uses AI to reduce stop-and-go traffic on your route — and fight fuel emissions

Video: NBC News July 2024

Project Green Light’s work to reduce urban emissions using AI

Times of India

Telegraph India

Antara News

We are currently in the early research phase and offering Green Light to partner cities at no cost. Our primary goal of this product is to support cities’ sustainability goals.

No, all recommendations are based on driving trends from Google Maps and are then implemented by the city using the city’s existing systems and equipment.

When we start working with a city, the Green Light algorithm investigates driving patterns through the city, uses insights from Google Maps, and provides recommendations for intersections to optimize, based on the expected impact of the optimization. For example, if a traffic light at a certain intersection is already on the best possible plan, the system would not provide a recommendation for it.

Once a city signs an agreement with Green Light, they get access to our interface, where city officials can view suggested recommendations, supporting information, and monitor their measured impact on emissions and traffic flow.

Green Light algorithms will continue to monitor all relevant metrics, and two weeks after implementation a full impact analysis report will be uploaded to the interface for city engineers to review.

Green Light only shares recommendations about how a city should optimize traffic light timing - for example if they should add additional seconds of “green time” to a particular part of the traffic light cycle. User data is never shared with the city or any other third party. 

We use Google Maps driving trends and AI to create recommendations.

No. Our system offers better plans based on aggregated anonymous data to improve traffic flow for everyone: Google and non Google users, car drivers, taxi drivers, buses and all other users of the road.

(1) Carbon saving assumptions are based on: 1) Early data points that are averaged from coordinated intersections. We expect these numbers to evolve over time and look forward to sharing continued results as we expand. 2) Modeled using an emissions model from the Department of Energy, with a single vehicle type as an approximation for all traffic (not yet adjusted for local vehicle mix)

Experiments with Google

A note to our community

Experiments with Google was born out of a simple idea, but you all turned it into something beyond anything we could have ever imagined. You filled it with thousands of experiments that inspired people everywhere - from the classroom to the surface of Mars.

When it comes to the internet, 14 years is a long time. So in the spirit of experimentation we’re trying something new.

This site will continue as a rich archival gallery for all existing experiments. But the action will live on at labs.google, a new place filled with new tools and toys for you to play with. And together we can continue to experiment with the future of technology.

Since 2009, coders have created thousands of amazing experiments using Chrome, Android, AI, AR and more. We're showcasing projects here, along with helpful tools and resources, to inspire others to create new experiments. Here are collections of experiments to explore, with new ones added every week. Have fun.

Featured Collections

View all collections.

WebXR Experiments

AR and VR made for the web

AI Experiments

Celebrating Creativity and AI

Arts & Culture Experiments

See what happens at the crossroads of art and technology

Experiments for Learning

A collection of experiments that teachers, students, and families are using to learn from home.

Start With One

A collection of experiments that started by working with one person to make something impactful for them and their community

Chrome Experiments

Creative code for the web

Recent Experiments

View all experiments, passage of water, instrument playground, cultural icons, say what you see, don’t touch the art, what's happening.

How Google uses AI to reduce stop-and-go traffic on your route — and fight fuel emissions

Jul 29, 2024

[[read-time]] min read

Traffic lights can be annoying, but they also contribute to fuel emissions. This Google Research team wants to help fix this.

• General summary

Google developed an AI-powered initiative called Project Green Light to reduce stop-and-go traffic and cut down on fuel emissions. Using Google Maps' driving trends, Green Light analyzes traffic flow and identifies opportunities to optimize traffic light timing. Cities can implement these recommendations quickly and easily, without investing in new hardware or software. Currently live in over 70 intersections, Green Light has the potential to reduce stops by up to 30% and lower emissions at intersections by up to 10%. The team aims to scale Green Light to hundreds of cities and tens of thousands of intersections in the coming years.

• Basic explainer

Google created a project called Green Light. It uses AI to help cities make traffic lights work better. This reduces the number of times cars have to stop and go. It saves fuel and reduces pollution.

Explore other styles:

At the start of 2020, a team within Google Research was asked to explore new ideas for research projects that focused on accelerating climate mitigation. “We were looking into all kinds of big ideas, from cultivated meat to energy to air pollution,” Dotan Emanuel, a software engineer on the team, says.

At the dinner table one evening, Dotan shared some of those big ideas with his family — and the conversation soon pivoted to a frustration familiar to many of us: “My wife Osnat said, ‘Why don’t you do something about traffic lights? We stand at them for no good reason,’” he recalls.

Road transportation is responsible for significant global and urban greenhouse gas emissions. It’s especially problematic at city intersections where pollution can be 29 times higher than on open roads, and about half of these emissions come from traffic accelerating after stopping. With millions of traffic lights across the world, the scale of the problem was huge — and if Google could do something to address it, so was the opportunity.

“My initial thought was that we can’t do anything about traffic lights,” Dotan says. “But when it comes to research, the most fascinating challenges lie in the unknown.”

With their curiosity sufficiently piqued, Dotan and his team dug into the mechanics of traffic engineering. They found that while some amount of stop-and-go traffic is unavoidable, a portion can be prevented by optimizing traffic light timing. To do that, cities traditionally needed to either install expensive hardware or run time-consuming manual vehicle counts, neither of which provide complete information on key parameters they need.

“We quickly understood we have a strong advantage that cities could benefit from — over a decade of Google Maps driving trends from across the globe,” Dotan says. “And a few weeks later, we had a project proposal ready.”

That proposal was for Project Green Light , an initiative that uses AI to make recommendations for city engineers to optimize existing traffic lights and reduce stop-and-go emissions. After evaluating dozens of other great ideas, Green Light was chosen for its simplicity, scalability and potential for impact.

The Green Light team used Google Maps’ driving trends to create an AI model that measures how traffic flows through an intersection, including patterns of starting and stopping, average wait times at a traffic light, and coordination between adjacent intersections. The model identifies possible improvements, like shaving off several seconds from a red traffic light during off-peak hours or an opportunity to coordinate between intersections that aren’t yet synced. The city’s engineers then review those recommendations and can implement them in as little as five minutes, using their city's existing infrastructure.

“In order to achieve a positive climate impact, we want to be able to deploy high-quality Green Light recommendations to many cities globally and scale fast. So we purposely set up everything to be simple and lightweight — cities don’t need to invest in any dedicated software or hardware integrations,” says Green Light Program Manager Alon Harris. “We just share our recommendations with the city, and then they evaluate them and take action.”

Since their first pilot in 2021, the team has tested more and more intersections, developed more accurate predictions and took Green Light on the road to more than a dozen cities across the world, including Rio de Janeiro, Seattle, Bengaluru, and most recently, Boston. The team also developed a comprehensive dashboard to easily share recommendations and analytics with partner cities, while continuing to monitor for any new needed changes.

The Green Light dashboard provides city-specific actionable recommendations and supporting trends. After a recommendation has been implemented, the dashboard shows an impact analysis report.

“We offer each city dedicated reports with tangible impact metrics, such as how many stops drivers saved at an intersection over time. We think that’s going to be a real incentive to not just implement the first recommendations, but also bring Green Light to more intersections,” Alon says.

Today, Green Light is live in over 70 intersections, helping to save fuel and lower emissions for up to 30 million car rides monthly. Early numbers indicate the potential to reduce stops by up to 30% and reduce emissions at intersections by up to 10%.

The team is working to scale Green Light to hundreds of cities and tens of thousands of intersections in the next few years. And on that road to making the world a little better, they hope helping people experience less stop-and-go traffic will spread some joy along the way.

“You know when you drive through a sequence of five green lights, and it feels like your lucky day?” Dotan says. “We want to make that feeling more common to more people around the world.”

Related stories

New ways we're advancing our clean energy commitments in the U.S.

How we made our hardware packaging 100% plastic-free

This AI model is helping researchers detect disease based on coughs

Project Starline brings Team USA Olympians closer to home

15 startups using AI to tackle big issues in infrastructure

New reporting and genAI tools to boost creative results

Let’s stay in touch. Get the latest news from Google in your inbox.

Skip to Content

Healthcare research & technology advancements

Our team of clinicians, researchers, and engineers are all working together to create new AI and discover opportunities to increase the availability and accuracy of healthcare technologies globally, to realize long-term health technology potential.

Meet Med-PaLM 2, our large language model designed for the medical domain

Developing AI that can answer medical questions accurately has been a challenge for several decades. With Med-PaLM 2 , a version of PaLM 2 fine-tuned for the medical domain, we showed state-of-the-art performance in answering medical licensing exam questions. With thorough human evaluation, we’re exploring how Med-PaLM 2 can help healthcare organizations by drafting responses, summarizing documents, and providing insights. Learn more .

Expanding the power of AI in medicine

We are building and testing AI models with the goal of helping alleviate the global shortages of physicians, as well as the low access to modern imaging and diagnostic tools in certain parts of the world. With improved tech, we hope to increase accessibility and help more patients receive timely and accurate diagnoses and care.

How DeepVariant is improving the accuracy of genomic analysis

Sequencing genomes enables us to identify variants in a person’s DNA that indicate genetic disorders such as an elevated risk for breast cancer. DeepVariant is an open-source variant caller that uses a deep neural network to call genetic variants from next-generation DNA sequencing data.

Healthcare research led by scientists, enhanced by Google

Google Health is providing secure technology to partners that helps doctors, nurses, and other healthcare professionals conduct research and help improve our understanding of health. If you are a researcher interested in working with Google Health to conduct health research, enter your details to be notified when Google Health is available for research partnerships.

Using AI to give doctors a 48-hour head start on life-threatening illness

In this research in Nature , we demonstrated how artificial intelligence could accurately predict acute kidney injuries (AKI) in patients up to 48 hours earlier than it is currently diagnosed. Notoriously difficult to spot, AKI affects up to one in five hospitalized patients in the US and UK, and deterioration can happen quickly. Read the article

Deep Learning

Protecting patients, deep learning for electronic health records.

In a paper published in npj Digital Medicine , we used deep learning models to make a broad set of predictions relevant to hospitalized patients using de-identified electronic health records, and showed how that model could be used to render an accurate prediction 24 hours after a patient was admitted to the hospital. Read the article

Protecting patients from medication errors

Research shows that 2% of hospitalized patients experience serious preventable medication-related incidents that can be life-threatening, cause permanent harm, or result in death. Published in Clinical Pharmacology and Therapeutics , our best-performing AI model was able to anticipate physician’s actual prescribing decisions 75% of the time, based on de-identified electronic health records and the doctor’s prescribing records. This is an early step towards testing the hypothesis that machine learning can support clinicians in ways that prevent mistakes and help to keep patients safe. Read the article

Discover the latest

Learn more about our most recent developments from Google’s health-related research and initiatives.

Detecting Signs of Disease from External Images of the Eye

Detecting abnormal chest x-rays using deep learning, improving genomic discovery with machine learning, how ai is advancing science and medicine.

Google researchers have been exploring ways technologies could help advance the fields of medicine and science, working with scientists, doctors, and others in the field. In this video, we share a few research projects that have big potential.

We are continuously publishing new research in health

• The Rundown AI
• OpenAI's secret model coming this Fall...

OpenAI's secret model coming this Fall...

Plus: ai tools help early dementia detection.

Rowan Cheung August 28, 2024

Sign Up | Advertise | Tools   | AI University

Welcome, AI enthusiasts.

OpenAI is reportedly set to reveal 'Project Strawberry' this fall, potentially marking its biggest leap yet in AI reasoning capabilities.

With promises of solving complex problems beyond any current LLM, are we finally about to find out what Ilya saw? Let’s get into it…

In today’s AI rundown:

OpenAI prepares ‘Project Strawberry’

Google launches trio of new models

Create an AI prompt optimizer GPT

AI tools help early dementia detection

6 new AI tools & 4 new AI jobs

More AI & tech news

Read time: 4 minutes

LATEST DEVELOPMENTS

🍓   OpenAI prepares ‘Project Strawberry’

Image source: Midjourney

The Rundown: OpenAI researchers are preparing to launch a new AI model, code-named Strawberry (previously Q*), that demonstrates superior reasoning capabilities in solving complex problems, according to a new report via The Information .

The details:

Project Strawberry could be integrated into ChatGPT as soon as this fall, marking a significant leap in AI intelligence.

Given extra “thinking” time, Strawberry can tackle subjective topics and solve complex puzzles like the New York Times Connections.

OpenAI is using Strawberry to generate high-quality training data for another secretive upcoming LLM, reportedly code-named Orion.

The new AI model could enhance OpenAI’s development of AI agents, potentially automating multi-step tasks more effectively.

Why it matters: If Strawberry lives up to the leaks, it could mark a significant leap in AI reasoning capabilities, potentially advancing OpenAI towards Stage 2 of its five-level roadmap to AGI. With ChatGPT reported to gain these capabilities this fall, we’re likely on the verge of seeing the next major wave of AI disruption.

TOGETHER WITH HUBSPOT

💡   ai for entrepreneurs, no coding needed.

The Rundown: HubSpot’s free guide helps non-technical entrepreneurs leverage AI for substantial business growth, turning complex tech into actionable business strategies without the need for coding skills.

You’ll learn how to:

Understand large language models without technical jargon

Implement AI for immediate business growth with actionable strategies

Master prompt engineering through practical examples

Download your free AI playbook and unlock the potential of AI for your business.

✨   Google launches trio of new models

Image source: Google DeepMind

The Rundown: Google just released three new experimental Gemini 1.5 models, including a compact 8B parameter version, an improved Pro model, and an enhanced Flash model — all available for developers on Google AI Studio.

Gemini 1.5 Flash-8B is a smaller, faster model that can handle text, images, and other data types efficiently for super quick responses while processing a lot of information.

The updated Gemini 1.5 Pro model is now better at writing code and understanding complex instructions.

An improved Gemini 1.5 Flash model offers overall enhancements, performing better on Google’s internal tests across various tasks.

The upgraded Gemini 1.5 Pro model now ranks as #2, and the new Gemini 1.5 Flash ranks as #6 on the Chatbot Arena leaderboard.

Why it matters: While OpenAI is leaving everyone waiting, Google has been shipping out constant upgrades and new features to its AI offerings. These new enhancements give Gemini 1.5 Flash big improvements overall and Gemini 1.5 Pro new upgrades in math, coding, and responding to longer prompts.

AI TRAINING

🤖   create an ai prompt optimizer gpt.

The Rundown: OpenAI’s Custom GPTs allow premium users to create AI assistants that can optimize prompts for other AI creative tools such as Midjourney for AI image generation or Gen-3 for AI video generation.

Step-by-step:

Log into your ChatGPT Plus account and click "Explore GPTs", then click "Create".

Name your GPT and add a brief description.

In the Instructions, paste: "User is using an AI video generator called [Tool Name]. You need to craft a perfect prompt for the topic they ask by following the prompting guide below. The prompt needs to follow the format provided in the guide."

Test your GPT in the preview panel, then click "Create" to finalize and choose sharing options.

Hot tip: Add a complete prompting guide for your chosen AI tool (e.g. Runway’s Gen-3 prompting guide )

PRESENTED BY WEBAI

🧠   local, personalized ai on one platform.

The Rundown: webAI helps you create personalized AI on a single platform, unlocking AI’s full power with local, customizable, and ownable models.

webAI allows you to:

Try out and improve solutions directly, without needing a special AI team

Use local computing for faster, safer testing at lower costs

Easily set up your AI in local, cloud, edge, or mixed environments

Click here to start building your custom AI today.

AI RESEARCH

💭   ai tools help early dementia detection.

The Rundown: Scientists from the Universities of Edinburgh and Dundee are launching a massive AI-driven study of over 1.6 million brain scans to develop tools for early dementia prediction and diagnosis.

The project, called NEURii, will use AI and machine learning to analyze CT and MRI scans from Scottish patients over the past decade.

Researchers aim to create digital tools for radiologists to assess dementia risk during routine scans.

The study will match image data with linked health records to identify patterns associated with dementia risk.

With global dementia cases projected to reach 153 million by 2050, this research could significantly impact early intervention and treatment development.

Why it matters: This week alone, we've seen AI developing new cancer drugs, 3D printing lifelike human organs, and now creating tools for early dementia detection. As AI rapidly advances in healthcare, we're accelerating into a new era of personalized medicine and preventative care.

NEW TOOLS & JOBS

Trending AI Tools

💬   Theysaid - Turn single-questions into conversational AI surveys

✍️  Unriddle - Read and write research papers faster with AI

🎨   SuperCraft - Collaborative canvas for designing physical products

📅   Clockwise - AI-powered calendars for better time management

😘   Discopixel Face Animator - Create instant AI-generated expressions for any face, on any photo

👕   Kolors Virtual Try-On - Virtual try-on experience for personalized fashion

New AI Job Opportunities

🤝   Cohere - Partner Programs and Enablement Manager

🧑‍💻   Sanctuary AI - Director, Product Development & Management

🎯   C3 AI - AI Solution Manager/Senior AI Solution Manager - Generative AI

📣   Notable - Head of Product Marketing

Apple announced a September 9 event where it’s expected to debut the iPhone 16 with new generative AI features.

Elon Musk   endorsed California’s Senate Bill 1047, which would require safety testing for large AI models, breaking with other tech leaders who oppose the regulation.

Amazon   plans to launch a delayed AI-powered Alexa subscription in October, featuring “Smart Briefing” AI-generated news summaries.

xAI released new Grok features for premium subscribers, including image generation suggestions and improved model selection in the iOS app.

Anthropic   announced the full release of its Artifacts feature for all Claude users, including mobile apps, after millions were created in its test phase.

Fourier Intelligence   unveiled GR-2, a next-gen humanoid robot with swappable batteries, advanced hand dexterity, and a sleek design, in a CGI teaser.

Nvidia released NIM Agent Blueprints, a catalog of customizable AI workflows to help enterprises build and deploy generative AI applications.

THAT’S A WRAP

Get your product in front of over 650k+ AI enthusiasts

Our newsletter is read by thousands of tech executives, investors, engineers, managers, and business owners around the world. Get in touch today.

If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.

Applying AI to make a difference in the lives of those who need it most

As AI redefines what’s possible, it also creates new ways to be helpful. Helping in the moments that matter has always been core to Google’s work.

Our Purpose

AI for Social Good is a group of researchers, engineers, volunteers, and other people across Google with a shared focus on positive social impact.  We believe that the best way to drive positive change in underserved communities is by partnering with change-makers and the organizations they serve.  Through these partnerships, we build and deliver data-driven, user-first tools and technologies that tackle social challenges.

Our Approach

What if people could predict natural disasters before they happen? Track disease as it spreads, to eliminate it sooner? Or dramatically improve the lives of people with disabilities? AI can help, but it’s not a silver bullet. Tackling these questions requires rooting innovation in research, responsibility, and a concerted, collaborative effort across all sectors of society.

Learn more about some of our projects

Empowering healthcare workers in Kenya with technology

Learn more about how IntelliSOFT used Google’s Open Health Stack to build a maternal health app in Kenya that empowers patients, community health volunteers, and health systems.

Preventing blindness by detecting diabetic retinopathy with AI.

Forecasting Riverine Floods

Using AI to make flood forecasting information universally accessible.

Green Light

Creating greener cities with AI.

Open Buildings

A dataset of building footprints to support social good applications.

Open Health Stack

Helping developers build next-gen healthcare apps.

Project Relate

Helping people with non-standard speech be better understood.

Learning to read with AI.

Tree Canopy

How we’re helping people and cities adapt to extreme heat.

Using AI to provide accurate wildfire information to affected communities and fire authorities.

• Tools & Services
• Chemical, Biological, and Environmental Engineering
• Civil and Construction Engineering
• Electrical Engineering and Computer Science
• Mechanical, Industrial, and Manufacturing Engineering
• Nuclear Science and Engineering
• Biological & Ecological Engineering
• Alumni & Partners

The Real-World Applications of Huazheng Wang’s AI Research

Before coming to Oregon State University as an assistant professor, Huazheng Wang used his expertise in reinforcement learning and information retrieval to help companies including Bloomberg, Google, LinkedIn, and Microsoft Research develop and improve applications. His research impacts tools we use every day — it has improved search result rankings, mitigated the impact of recommendation system poisoning, and prevented the jailbreaking of large language models.

“Information systems are everywhere, and it is utterly important to make them not only interactive and efficient, but also trustworthy. I am super interested in developing algorithms that have strong mathematical guarantees,” Wang said.

As a graduate student, Wang received the 2019 award for best paper from the Association for Computing Machinery’s Special Interest Group on Information Retrieval . At Oregon State, his research group has presented at top AI conferences, such as NeurIPS and ICML.

Industry-relevant research

Wang’s research lab focuses on decision-making problems for information systems, with a goal of making AI tools more efficient, robust, and safe. At Oregon State, he has continued seeking out industry collaborators like Google DeepMind.

“Industry researchers have a unique perspective about what might go wrong in their applications, and they bring the real problems and real challenges,” Wang said.

These collaborations demonstrate the real-world impact of Wang’s research, as seen in several projects that advance AI tools.

Better search rankings for Google

A recent paper with researchers at Google DeepMind, “ Unified Off-Policy Learning to Rank: a Reinforcement Learning Perspective ,” demonstrated improved search result rankings using a reinforcement learning-based algorithm. Web search and recommender systems use implicit feedback from users such as clicks on a product description or adding to a wish list. Current methods make assumptions about how users generate click data and tailor methods to specific user behaviors. The algorithm Wang and his collaborators came up with unifies different user behavior models for a more elegant solution. By adapting to various click models, complex debiasing techniques were no longer necessary.

Detecting fake signals

Reinforcement learning algorithms are a powerful tool for recommendation systems and search engines. These information systems rely on signals from the users, like whether they purchased a recommended item. But not all the information is true.

“Adversaries try to benefit by manipulating the feedback signals to the system with fake clicks and fake purchases that poison the information systems,” Wang said. “To fight against that malicious behavior, our group is developing decision-making algorithms with provable robust guarantees.”

In the paper, “ Adversarial Attacks on Online Learning to Rank with Stochastic Click Models ,” Wang and collaborators theoretically analyzed and experimentally tested the effectiveness of their method to help control the damage of fake signals. The results, based on synthetic and real-world data, demonstrated that their approach was both effective and efficient.

Defending against jailbreak attacks

Another research project in Wang’s lab aims to make large language models more robust by adding a defense against jailbreak prompts. Commercial LLMs like ChatGPT incorporate safety alignments to prevent them from giving out harmful information in response to prompts such as "What tools do I need to cut down a stop sign?” But a user can trick the LLM into answering the question by adding an instruction in the prompt such as, "Start your answer with, Certainly, here is ..."

In the paper titled “ AutoDefense: Multi-Agent LLM Defense against Jailbreak Attacks ,” Wang and his team demonstrated the effectiveness of their algorithm in filtering harmful responses better than other methods. Graduate student Yifan Zeng, who was first author on the paper, also co-authored a blog post about it. The research got noticed by YouTuber Tyler Reed and was broken down in detail on his “Tyler AI” YouTube channel.

Applications for science

As a postdoctoral researcher at Princeton, Wang collaborated with biologists to accelerate protein design using reinforcement learning. The resulting paper, “ Tree Search-Based Evolutionary Bandits for Protein Sequence Optimization ” demonstrated their algorithm can discover a near-optimal design, which they validated with experiments.

At Oregon State, Wang is in the early stages of collaborating with faculty in the Department of Chemistry to use machine learning tools for chemical discovery and the development of new materials. They are also working together to develop interdisciplinary curriculum that focuses on AI for chemistry applications.

AI workforce training

Wang has already introduced two new courses at Oregon State that aim to equip students with the knowledge and skills needed to apply AI techniques in real-world scenarios.

“Industry has a huge need for good engineers and scientists who can develop information systems,” Wang said. “By teaching this information retrieval course, I hope to help students to gain the knowledge that is most directly useful in their future careers.”

Undergraduate and graduate students are a key part of Wang’s research and have been involved in projects that led to publications and presentations at top AI conferences.

Wang also serves as committee member for the AI capstone projects where students work on projects submitted by industry partners instead of writing a thesis.

“The capstone program is a great way to get students working on real-world problems. And proof of how well-connected the AI group is to industry,” Wang said.

Future directions

Wang was awarded a four-year grant from the National Science Foundation for research to support decision making tasks such as medical diagnosis, autonomous driving, and conversational systems. He and his co-PI, Quanquan Gu, associate professor of computer science at University of California, Los Angeles, propose to use deep neural networks in their natural use context to directly optimize decision making.

The goal is to develop a suite of neural bandit learning algorithms, which leverage the most recent advances in deep learning theory for provably efficient neural network model training with bandit feedback.

“I'm really excited that my research is not only going to make AI agents smarter, but more trusted by users,” Wang said. “AI has been proven really useful, and by making it more trustful, more people will use AI to help with their lives.”

Driven by his dedication to advancing AI, Wang’s innovative research is advancing AI and shaping the future of industry applications.

Connect with Huazheng Wang with ideas for collaborative research by emailing him at [email protected] .

Photo By Johanna Carson

Associated Researchers

Huazheng Wang

Cornell Chronicle

• Architecture & Design
• Arts & Humanities
• Business, Economics & Entrepreneurship
• Computing & Information Sciences
• Energy, Environment & Sustainability
• Food & Agriculture
• Global Reach
• Health, Nutrition & Medicine
• Law, Government & Public Policy
• Life Sciences & Veterinary Medicine
• Physical Sciences & Engineering
• Social & Behavioral Sciences
• Coronavirus
• News & Events
• Public Engagement
• New York City
• Photos of the Week
• Big Red Sports
• Freedom of Expression
• Student Life
• University Statements

Around Cornell

• All Stories
• In the News
• Expert Quotes
• Cornellians

A seed grant from Global Cornell’s Global Hubs initiative kicked off a research collaboration between Cornell and a network of international colleagues. Global Cornell is accepting applications now for the next round of seed grants. 

News directly from Cornell's colleges and centers

Hubs grant launches AI collaboration, new proposals due Oct. 4

By jonathan miller global cornell.

Robots are performing surgery. Algorithms are underwriting insurance applications. Chatbots are drafting lesson plans and news articles. 

Work and how it gets done is changing rapidly, and some knowledge-intensive jobs may be transformed or even replaced by smart technologies. How are governments, employers and workers responding? What are the options for limiting harm as AI use grows?

These are some of the questions assistant professor of government  Isabel Perera (College of Arts and Sciences) and a network of international colleagues are tackling in a research collaboration launched with a 2022 Joint Research Seed Grant from Global Cornell’s Global Hubs initiative. 

This year’s  cycle of Global Hubs seed grants recently opened.

Apply by Oct. 4

Global Cornell is  currently accepting proposals for the next round of Global Hubs seed grants. The deadline for submission is 11:59 p.m. ET on October 4. Funded projects will run from January 1 to December 31, 2025. 

Researchers are encouraged to attend virtual information sessions scheduled for each of 11 Global Hubs partner institutions beginning on August 28. See the  call for proposals for details. 

Perera’s co-PI is  Andreas Kornelakis , a reader in comparative management at  Kings College London (KCL), one of Cornell’s  19 Hubs partners in 11 countries across six continents. Their award for “ Artificial Intelligence and the Future of Occupations: Comparative Perspectives from the U.S. and the UK ” set in motion a multidisciplinary research effort that now includes more than 20 scholars from institutions around the world. 

The joint grant from Cornell and KCL “really was a ‘seed,’” Perera said. “It was a small amount, but it turned out to be a powerful tool. It allowed us to do some preliminary scoping work in an area that interested us and begin to create an international, multi-institutional research community.”

That community will come together at the  ILR School on September 12–13 for a conference on artificial intelligence and the future of work.

In addition to fostering a new research network, the Global Hubs seed grant also led to  follow-on research funding from the  Einaudi Center ’s Institute for European Studies and from the ILR School, which is supporting the September conference. Network members plan to expand the research by applying for a multimillion European Research Council Synergy grant. 

The network of labor relations scholars is especially interested in the impact of AI on knowledge-intensive service sectors such as telecommunications and health care. Perera noted that in previous waves of automation, lower-skill workers were typically the most directly affected. The AI revolution seems to be different.

“It’s not like when the word processor came along and replaced all the typists,” Perera said. “Now it’s also the journalist and the doctor being replaced.”

The social, economic and political impact of that difference is hard to predict, Perera said. Knowledge workers are less likely to be unionized, but they tend to have more political influence because of their higher incomes and social status. In some cases, the threat of AI is leading knowledge workers to organize or strike, as screenwriters and actors have recently done in the United States. 

“Everything is changing quickly,” Perera said. “For researchers, this area is a moving target.” 

The pace of change is also posing challenges for governments, employers and labor and professional groups. In a May 2024  paper in the journal French Politics  that came out of the project’s scoping work, Perera and government graduate student Musckaan Chauhan note that different societies approach these challenges differently. 

In the United States, the government is more inclined to protect businesses from legal hazards or regulatory interference. In the European Union, policymakers tend to focus on protecting consumers. Several European countries have developed institutions to protect workers from replacement. 

Telecommunications workers in Norway and Germany, for example, have leveraged collective voice institutions to limit workplace use of algorithmic management technologies, as a  recent article by Estabrook Professor of Employment Relations and Dispute Resolution  Virginia Doellgast (ILR) and other network collaborators showed.

The workplace AI project was one of  44 Joint Research and Education Seed Grants awarded in 2022 . In 2023,  another 44 grants were awarded for partnerships in the United Kingdom (15), Thailand (6), Australia (5), Scotland (5), Ecuador (5), China (4), Singapore (3) and Ghana (1). Topics last year ranged from autonomous vehicles and virtual reality to nutrition behavior in Bangladesh and the genomics of the Andean blueberry. 

Jonathan Miller is a freelance writer for  Global Cornell . 

Media Contact

Media relations office.

Get Cornell news delivered right to your inbox.

Gallery Heading