On Wednesday, May 10th, Google held its I/O conference, an opportunity for Google to tell more about its different projects. During this annual meeting, Google took the opportunity to unveil its new artificial intelligence model, PaLM 2.
What is PaLM 2?
In a talk show presented by Sundar Pichai, CEO of Google, PaLM 2 is an improved version of Google’s previous AI model, PaLM. Like Open AI’s GPT-4 or Meta’s LLaMA, PaLM 2 is already integrated into Google Bard and offers many features.
Wishing to make artificial intelligence accessible to all, PaLM 2 has 4 modes: Gecko, Otter, Bison and Unicorn. Very little information was communicated about these modes. We only know that Gecko is a mode for smartphones and can be used even offline.
What are the capabilities of PaLM 2?
Developed to answer any problem, PaLM 2 was trained for several years on scientific and mathematical datasets. These same datasets also allow it to speak 100 languages, including Korean and Japanese. In an example given by Pichai, PaLM 2 could debug code in Japanese if a user asked it.
Besides very advanced general knowledge, PaLM 2 has several variants. These variants provide it with much more knowledge depending on the chosen domain. Thanks to a variant called “fine-tuning” it can answer questions and problems of cybersecurity like detecting threats or malicious scripts. Another example is medicine with “Med-PaLM 2” which allows it to interpret an X-ray image.
At the end of the presentation, PaLM 2 was announced as already available in a pre-release. A final statement by Pichai said that this program is already integrated in 25 Google products, including Bard. In any case, if you enjoyed this article and want to learn more about the Google I/O conference, or if you’re considering a career in Data Science, feel free to check out our articles or training offers on DataScientest.
Source : io.google
