Google has just unveiled PaLM 2, its next-generation artificial intelligence language models, which already power 25 of the company's products, including Bard. This announcement promises to revolutionize AI and give the technology a boost.
PaLM 2, the successor to PaLM, already powers several Google products
Google's next-generation artificial intelligence language model, PaLM 2, was introduced at Google I/O 2023. This revolutionary technology is already powering 25 of the company's products, including Bard. The successor to PaLM, introduced in 2022, this language model comes in different sizes and names: Gecko, Otter, Bison and Unicorn. The smallest of these, Gecko, is even compact enough to run on a smartphone.
PaLM 2 is compatible with over 100 languages and is also capable of writing code. The demonstration showed that the AI was able to identify errors in a block of code and add comments in Korean to each line. Thus, Bard has recently integrated this feature to help with programming.
Google DeepMind: a promising merger for AI development
PaLM 2 is the joint work of two teams: Google Brain and DeepMind. These two entities have recently united under the name Google DeepMind and are currently working on Gemini, their next-generation founding model. Details on Gemini are still scarce, but Google is hinting at major advances in the field of AI.
Improvements in logic, reasoning, and math with PaLM 2
According to Google, the dataset used to train PaLM 2 includes scientific documents and websites containing mathematical expressions. The company claims that its new language model has “improved logic, common sense reasoning and mathematical capabilities” compared to its predecessor.
PaLM 2 in the service of medicine and cybersecurity
PaLM 2 is also used to enhance Bard's code creation, interpretation, and debugging capabilities. It also supports chatbot in Japanese and Korean and extends its functionality to Workspace. In addition, Google highlighted the specific use of PaLM 2 in areas such as medicine and cybersecurity.
Med-PaLM 2 was introduced last April as part of an effort to “more accurately and safely answer medical questions” via its platforms. In the demonstration, the AI was able to interpret a lesion from an X-ray image.
On the other hand, Sec-PaLM was specifically mentioned. It is a model designed for use cases related to computer security, helping experts to fight threats.
PaLM 2 preview: a promising future for generative AI
PaLM 2 is available as a preview today and promises to bring great news to the generative AI race in the coming months. Interesting examples of developments using the PaLM API were also presented, both at the commercial and experimental level.
Among these developments, a prototype called Project Tailwind was shown, capable of generating an AI model based on documents stored in Google Drive by a user. The demonstration highlighted its potential as a tool for students, capable of generating summaries or glossaries. Its application could be even broader, but it remains to be seen whether this will evolve into a large-scale product or tool.
The announcement of PaLM 2 and Gemini shows that Google is ready to take on the GPT-4 challenge and push the boundaries of artificial intelligence. The progress Google DeepMind is making is promising and could transform the future of AI and technology in general.