Neural networks were introduced to me as an undergrad. Many people in the community were excited about the potential of neural networks but couldn't yet accomplish important tasks. I was happy as well. I used parallel computation to train neural networks and thought we only needed 32X more compute power. I wasn't very far off. We needed 1 million times more computation power.
It was time to take another crack at neural networks 21 years after the first one. Neural networks were trained using millions of randomly selected frames from online videos. The results were amazing. The system learns to recognize cats without explicit training. This was one of the most significant discoveries in the field of artificial intelligence, which is still being made at many places.
I share my history of neural networks to show that progress in artificial intelligence has been gradual. Prior to 2012 computers were difficult to see, hear, and understand. We have made rapid progress in artificial intelligence over the past decade.
We are excited about the recent advances in artificial intelligence that are being led by the company and are being used to help people around the world. In many fields of science and other human endeavors, we need to deploy artificial intelligence in the cloud.
We are aware of the risks associated with the technology. The first major company to release and operationalize a set of artificial intelligence principles has allowed us to focus on making rapid progress on technologies that can be helpful to everyone. It needs to be a collective effort, involving not just researchers, but domain experts, developers, community members and businesses.
I am happy to announce that I will use artificial intelligence to make technology accessible in many more languages. Exploring how artificial intelligence might boost creativity. Climate adaptation is included in the third in the artificial intelligence for social good.
People use language to communicate and understand the world. It is the most natural way to use technology. More than 7,000 languages are spoken around the world and only a few are represented online today. Traditional approaches to training language models on text from the web don't capture the diversity of how we communicate around the world. The pursuit of our mission to make the world's information universally accessible and useful has been hampered by this.
The 1,000 Languages Initiative is an ambitious commitment to build an artificial intelligence model that will support the 1,000 most spoken languages and bring greater inclusion to billions of people in marginalized communities. Some may even call it a moonshot, but we are already making progress and seeing the path clearly. The way people use technology has changed a lot. Images, videos, and speech are some of the new methods people are using to find and share information. Multiplying means that our most advanced language models are able to open information in many different formats. New opportunities arise from these shifts.
The Universal Speech Model, or USM, is the largest language coverage seen in a speech model to date, training on over 400 languages. We are working with communities across the world to source speech data. In order to create and publish data in Africa, we recently announced voice typing for 9 more African languages on Gboard. In South Asia, we are working with local governments, NGOs, and academic institutions to collect audio samples from across the region.
People in different cultures can use video, imagery, and design in ways that they previously couldn't with the help of generative models.
Our researchers have been working hard to create models that human raters prefer over other models. Applying our model to video sequence and generating long coherent videos for a sequence of text prompt was a recent breakthrough. For the first time, we can combine these techniques to make video.
People will be able to learn about, experience, and give feedback on emerging artificial intelligence technology when we bring our text-to-image generation technologies to the artificial intelligence test kitchen. Users will be able to give feedback on the demos in the second season of the show. You will be able to build themed cities with "City Dreamer" and design friendly monster characters that can move, dance, and jump with text prompt.
Text-to-3D is a reality with DreamFusion, which creates a three-dimensional model that can be viewed from any angle and can be used to create any 3D environment. Audiolm is a model that learns to generate realistic speech and piano music by listening to audio only. AudioLM can predict which sounds will follow after a few seconds of an audio prompt by using a language model similar to that of a language model.
As we develop these tools, we're working with creative communities around the world. Wordcraft is built on our state-of-the-art dialog system LaMDA and is being used to experiment with artificial intelligence. The first volume of these stories can be read at the Wordcraft Writers workshop.
The effects of climate change can be addressed with the help of artificial intelligence. Hundreds of thousands of people are affected by wildfires today and are increasing in scale.
Satellite imagery has been used to train artificial intelligence models to identify and track wildfires in real time, helping predict how they will evolve and spread. We launched this wildfire tracking system in the U.S., Canada, Mexico, and are currently rolling out in parts of Australia, and since July we have covered more than 30 big wildfire events in the U.S.
Floods are one of the extreme weather patterns caused by Climate change. We helped save many lives by sending 115 million flood alert notifications to 23 million people over the internet. We are now covering more countries in South America and Sub-Saharan Africa. Transfer learning is an artificial intelligence technique that works in areas where there is less data. The global launch of FloodHub is a new platform that displays when and where floods can occur. More people will be able to reach safety in flooding situations when this information is brought to search and maps by us.
More access to healthcare is being provided by Artificial Intelligence. We are researching ways to use artificial intelligence to give parents the information they need to identify issues early in a pregnant woman's life. We plan to work with caregivers and public health agencies to expand access to diabetes screening through the ARSA. More than 150,000 patients have been successfully screened through ARDA in countries like India, Thailand, Germany, the United States, and the UK. We are looking at how artificial intelligence can help detect respiratory and heart rates. Making healthcare more accessible for anyone with a phone is one of the goals of this work.
Neural network architectures, machine learning and new approaches to hardware for machine learning have helped artificial intelligence solve real world problems for billions of people. There is much more to be done. We are sharing a hopeful vision for the future with the help of artificial intelligence. We hope to use this technology to improve people's lives around the world.
All Rights Reserved © 2024
