At the Google I/O 2023, Google announced a host of new features and services that will take advantage of generative AI. Along with it, the company also unveiled its latest iteration of Large Language Model. Called PaLM 2, the new LLM is built upon the foundation of the existing model and is already powering 25 Google services including the company"s chatbot Bard.
At the event, Google shared the improvements that PaLM 2 brings to the table. These include improved multilingual capabilities and is trained in over 100 languages. Google further claims that its LLM can pass "advanced language proficiency exams at the “mastery” level." Moreover, PaLM 2 is also trained on a variety of scientific papers, journals and websites giving it great research and mathematics capabilities. Lastly, the model has also been trained on a large amount of codebase making it proficient in not only modern languages like Python and JavaScript but also languages like Prolog, Fortran and Verilog.
Google has already deployed PaLM 2 within 25 of its services which includes productivity tools like Google Docs, Sheets and Gmail. Furthermore, the model is also powering specialized Google services like Med-PaLM 2 which has achieved "state-of-the-art results in medical competency, and was the first large language model to perform at “expert” level on U.S. Medical Licensing Exam-style questions." Other specialized examples include Sec-PaLM which used PaLM 2 and Google Cloud to scan malicious scripts, cyber threats and more.
At the event, Google stated:
Even as PaLM 2 is more capable, it’s also faster and more efficient than previous models — and it comes in a variety of sizes, which makes it easy to deploy for a wide range of use cases. We’ll be making PaLM 2 available in four sizes from smallest to largest: Gecko, Otter, Bison and Unicorn. Gecko is so lightweight that it can work on mobile devices and is fast enough for great interactive applications on-device, even when offline. This versatility means PaLM 2 can be fine-tuned to support entire classes of products in more ways, to help more people.
At last, Google also teased Gemini, a new multimodal that is highly efficient at tool and API integrations. Google notes that Gemini is currently getting trained and once finely tuned, it will be available "at various sizes and capabilities, just like PaLM 2, to ensure it can be deployed across different products, applications, and devices for everyone’s benefit."