Today, Google unveiled PaLM 2, their latest state-of-the-art language model, during I/O 2023. With improved multilingual, reasoning, and coding abilities, PaLM 2 is set to revolutionize the AI landscape.
Multilingual Prowess
PaLM 2 has undergone extensive training on multilingual text, covering over 100 languages. This has greatly enhanced its capacity to understand, generate, and translate complex text, including idioms, poems, and riddles across numerous languages. Moreover, PaLM 2 has achieved “mastery” level on advanced language proficiency exams.
Enhanced Reasoning
Thanks to its diverse dataset featuring scientific papers and web pages containing mathematical expressions, PaLM 2 demonstrates improved logic, common sense reasoning, and mathematics capabilities.
Coding Expertise
PaLM 2 was pre-trained on an extensive collection of publicly available source code datasets, which allows it to excel at popular programming languages like Python and JavaScript, as well as generate specialized code in languages such as Prolog, Fortran, and Verilog.
A Versatile Range of Models
PaLM 2 offers a variety of sizes to cater to different use cases, including Gecko, Otter, Bison, and Unicorn. The smallest, Gecko, is lightweight enough to operate on mobile devices and provides swift, interactive applications even when offline. This versatility enables PaLM 2 to support a wide array of products and cater to diverse user needs.
Powering Over 25 Google Products and Features
Over 25 new products and features powered by PaLM 2 were announced during I/O, bringing cutting-edge AI capabilities to consumers, developers, and enterprises worldwide. Some notable examples include:
- Expanding Bard to new languages with PaLM 2’s improved multilingual abilities and powering the recently announced coding update.
- Enhancing Workspace features in Gmail, Google Docs, and Google Sheets, all tapping into PaLM 2’s capabilities for faster and more efficient task completion.
- Med-PaLM 2, a specialized version trained with medical knowledge, can answer questions and summarize insights from dense medical texts, achieving state-of-the-art results in medical competency and expert-level performance on U.S. Medical Licensing Exam-style questions. It is also being developed with multimodal capabilities to synthesize information from sources like x-rays and mammograms to improve patient outcomes. Med-PaLM 2 will be available to a small group of Cloud customers for feedback later this summer.
- Sec-PaLM, another specialized version of PaLM 2, focuses on security use cases and promises a breakthrough in cybersecurity analysis. Available through Google Cloud, it uses AI to analyze and explain the behavior of potentially malicious scripts, rapidly detecting threats to people and organizations.
- Developers can now sign up to use the PaLM 2 model or access it through Vertex AI, featuring enterprise-grade privacy, security, and governance. PaLM 2 also powers Duet AI for Google Cloud, a generative AI collaborator designed to help users learn, build, and operate faster than ever before.
- Advancing the Future of AI
- PaLM 2 demonstrates the potential of versatile AI models in various sizes and speeds, offering benefits for everyone. Google is committed to creating the best foundation models while working on the next-generation AI model, Gemini. Designed to be multimodal, highly efficient, and capable of enabling future innovations like memory and planning, Gemini is currently in training and showcases never-before-seen multimodal capabilities. Once fine-tuned and rigorously tested for safety, Gemini will be available in different sizes and capabilities, similar to PaLM 2, ensuring its deployment across various products, applications, and devices to benefit a wide range of users.
- Google’s Brain and DeepMind research teams have achieved numerous defining moments in AI over the last decade. To accelerate progress, these two world-class teams are now merging into a single unit, Google DeepMind. Backed by Google’s computational resources, Google DeepMind will not only bring incredible new capabilities to the products used daily but also responsibly pave the way for the next generation of AI models.