PaLM-2 is Google’s most powerful language model to date and is expected to power many of the company’s AI services. Google also confirms the existence of Project Gemini.
Google unveiled PaLM-2 at Google I/O, calling it the next-generation language model. Compared to its predecessor PaLM, it is said to be able to speak more languages, reason better, and code better. PaLM-2 has been trained on texts from more than 100 different languages, numerous scientific publications, mathematical websites, and code databases.
The smallest PaLM-2 model is expected to run on mobile devices
There are no exact details about the size, but according to Google, the model will be available in four different sizes: Gecko, Otter, Bison and Unicorn. The smallest, Gecko, will even be able to run on mobile devices, enabling interactive applications on the device – even when the device is offline.
PaLM-2 also powers Google’s Bard chatbot, as well as AI features in workspace applications such as Gmail and Google Docs. Google’s language model is also available in two specialized variants: Med-PaLM-2 for medicine and Sec-PaLM for cybersecurity applications. The Med-PaLM-2 model will add multimodal capabilities, such as the ability to directly evaluate x-rays.
PaLM-2 and the special models are available – with restrictions – via Google’s PaLM API.
“Project Gemini” is already in development
With the announcement of PaLM-2, Google also commented on the merger of the Brain and Deepmind teams and rumors of “Project Gemini”. The team is responsibly paving “the way for the next generation of AI models” and is already working on Gemini, Google’s next multimodal model.
Gemini is multimodal from the ground up, highly efficient at integrating tools and APIs, and designed for future innovations such as memory and planning, Google said. Gemini is still in training, but according to the announcement, it already shows multimodal capabilities not seen in previous models.
“Once fine-tuned and rigorously tested for safety, Gemini 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,” Google said.