We aim to establish TensorFlow as a common platform for expressing various machine learning ideas, ensuring it can be used for deep learning, perception problems, and language understanding.

We conduct long-term research independent of any particular application and build open-source systems to assist with our research and deployment of machine learning models, like TensorFlow.

TensorFlow is our second-generation system designed to tackle deep learning and machine learning problems, built to be an open-source platform for global collaboration.

The current batch has a notable increase in developer tools, with about 30% more than four years ago, indicating a growing interest in building tools for AI development.

The new translation system significantly improved quality, reducing the codebase from 500,000 lines to just 500 lines of TensorFlow code, while achieving near-human translation quality.

Google is committed to both TPUs and GPUs, leveraging both for training and serving AI models, while maintaining flexibility in their infrastructure.

Open source AI development is progressing faster than proprietary models, as it allows for a diverse range of ideas and contributions from around the world.

Google Cloud will offer a virtual machine with a 180 teraflop TPU version 2 device, running TensorFlow programs quickly.

We believe that open-source models will drive more progress in AI, as contributions will come from various labs, not just the largest ones.