Google has been at the forefront of AI advancements, as highlighted by Zoubin Ghahramani, Vice President of Google DeepMind, in a recent blog post. He emphasized that the company's foundation models serve as the "bedrock for the industry and the AI-powered products that billions of people use daily."
During a presentation at the Shoreline Amphitheater, Ghahramani and other Google executives unveiled the latest version of their large language model, PaLM 2. Available in four sizes, PaLM 2 can operate locally on devices ranging from mobile phones to server farms. As a successor to the original PaLM model, which powered Google's experimental Bard AI, PaLM 2 demonstrates enhanced functionality.
Ghahramani described PaLM as a versatile model that can be fine-tuned for specific tasks. For instance, health research teams have customized PaLM to analyze medical texts, answering questions and summarizing complex information. Notably, PaLM was the first large language model to achieve expert-level performance on the U.S. medical licensing exam.
Bard now utilizes PaLM 2, benefiting from improved multilingual, reasoning, and coding capabilities. The updated model has been extensively trained on multilingual texts, encompassing over 100 languages with a better understanding of cultural nuances. Additionally, it excels at generating code in languages like Python and JavaScript. Its capabilities in logic, common sense reasoning, and mathematics have also been heightened through the use of diverse training data, including scientific papers.
Moreover, Google has developed application-specific versions of the base PaLM system, named Gecko, Otter, Bison, and Unicorn. "We built PaLM to be smaller, faster, and more efficient, while increasing its capabilities," Ghahramani stated. "This has allowed us to create a family of models in a range of sizes, with the lightest model able to run as an interactive application on the latest mobile devices."
In total, Google is announcing over two dozen products featuring PaLM capabilities at the I/O event, showcasing the company's continued innovation in AI technology.