PaLM 2 is a large language model developed by Google, serving as the successor to their previous state-of-the-art language model PaLM. PaLM 2 was designed to excel at advanced reasoning tasks, including code and math, classification, question answering, multilingual proficiency, and natural language generation, surpassing its predecessor in these areas. Its improved performance can be attributed to a combination of three research advancements in large language models, including the use of compute-optimal scaling, a more diverse pre-training dataset mixture, and updated model architecture and objectives. PaLM 2 was evaluated rigorously for the potential harms and biases it could cause, and its downstream uses for research and in-product applications. It is grounded in Google’s principles of responsible AI development and commitment to safety.PaLM 2 also demonstrates improved multilingual capabilities and has been pre-trained on a large quantity of webpage, source code, and other datasets, making it capable of coding in popular programming languages like Python and JavaScript, as well as generating specialized code in languages like Prolog, Fortran, and Verilog. PaLM2’s improved understanding of nuances in human language enables it to excel at understanding idioms and riddles, requiring a comprehension of ambiguous and figurative meanings. PaLM 2 is contributing to Google’s generative AI features and tools such as Bard, a tool that aids in creative writing and productivity, and the PaLM API, which provides a platform for developing generative AI applications. PaLM 2’s contributions to generative AI features and research are grounded in Google’s approach to building and deploying AI responsibly.
