Google began the keynote speech at its developer conference this year I/O 2023 by officially announcing PaLM 2, which is the second generation of the language artificial intelligence model on which the chatbot Bard relies ChatGPT competitor, which it also said now supports 25 of its services, and the keynote at the conference saw Google’s focus on AI technologies and its reliance on them in its products over the past years and its future plans to integrate more features depending on them.
What is PaLM 2?
Table of Contents
Google said that the new generation of its PaLM 2 language AI model provides high capabilities in a wide range of tasks, based on the latest research in AI technologies and the company’s infrastructure, and said that more than 25 of its products and features are now supported by the capabilities of the new PaLM2 model.
Google provides different models of PaLM 2 in different sizes that Google calls Gecko, Otter, Bison and Unicorn, and the company said that Gecko is so small that it can work on mobile devices, and it is fast enough for interactive applications on the device even when offline.
Google also said that PaLM 2 models are stronger in logic and reasoning thanks to extensive training on scientific and mathematical topics, and are trained in multilingual word processing (up to more than 100 languages).
In addition to the great ability to process code and help programmers in various programming tasks, Google said that PaLM 2 is also able to help developers around the world collaborate to write code, where for example a developer can ask his Korean colleague who only speaks Korean to help fix a bug by asking to add comments in Korean to the code produced by the model.
What are the features of Google’s PaLM 2 AI model?
PaLM 2 (Google’s new language model) offers much better capabilities and features compared to the first generation, most notably:
Support multiple languages
PaLM 2 is trained in multilingual word processing up to more than 100 different languages including Arabic, which makes it able to accurately understand, generate and translate texts, including idioms, poems and riddles in various languages.
Better abilities to understand logic, reasoning, and mathematics
Based on practicing a wide range of scientific papers and web pages that include mathematical equations, PaLM 2 offers better abilities in logic, reasoning, and mathematics.
Greater skills in helping programmers
Google said that PaLM 2 is already trained on a large amount of code that makes it able to help efficiently in the most famous programming languages such as Python, JavaScript and others, with support for creating specialized code in languages such as Prolog, Fortran and Verilog.
25 Google services and features powered by PaLM2 capabilities
Google said that more than 25 features and services from Google now rely on the capabilities of the new generation of its PaLM 2 language model, including:
-
- Bard chatbot that is now capable of better word processing, supporting more languages, and helping more with various programming tasks.
- Many smart features in Google’s Workspace desktop apps, including Help Me Write in both Google Docs and Gmail.
- Med-PaLM 2: It is the technology trained by Google’s health research teams, so that it can answer questions and summarize ideas from a variety of dense medical texts, which Google said is the first language model of artificial intelligence capable of exceeding the medical licensing test in the United States at the expert level, which it said it is working to enhance with the ability to extract information from X-rays and mammography to improve patient outcomes in the future, and the medical model for testing is available to a small group of customers Google Cloud later this summer 2023.
- Sec-PaLM: A specialized version trained in cybersecurity cases, through the use of artificial intelligence to help analyze and explain the behavior of potentially malicious software, as well as better detect software that may threaten individuals and organizations in an unprecedented time.
Google also announced three new artificial intelligence models for its cloud service customers Google Cloud besides PalM 2, including the Imagen model for creating images, in addition to the Codey model to help develop applications by creating code, along with the Chirp model for converting voice to text that supports 100 languages, including Arabic, and said that it is now working on developing the Gemini model to be multimedia and highly efficient.
