Google announced a new AI language model, PaLM 2, the second version of its PaLM language model that will be used to power Google Bard, at its I / O 2023 developer conference early this morning.
The PaLM 2 model is described as shining when fine-tuned for domain-specific data. For example, Med-PaLM 2 can help answer questions and summarize insights from a variety of dense medical texts.
In addition, Google recently moved the Bard chatbot to PaLM 2, claiming that with improved math, logic, and reasoning skills, Bard can now help generate, interpret, and debug code in more than 20 programming languages.
Google says that starting next week, the Bard chatbot will be more precise in its code references. If Bard introduces a piece of code, simply click on a comment and Bard will underline the block of code and link to the source code.
Google will soon launch extensions in Bard that leverage the entire Google service and others to help users do more work, such as Adobe, Spotify, YouTube, and more.
Notably, Google says the lightest version of PaLM 2, Gecko, is small enough to run on a phone and can process 20 tokens per second — the equivalent of roughly 16 or 17 English words. Google didn’t say what hardware it used to test the model, only that it ran “on the latest phones.
Google said it will launch more than 25 products and features powered by PaLM 2.