The LLM has been trained on over 20 programming languages - including some more specialized ones like Prolog and Fortran - so that it can write and debug code for more users. It can even explain steps and provide diagrams, meaning it can show its work if its math teacher ever asks.įinally, PaLM 2 is getting a coding upgrade.

It has been trained on mathematical expressions and scientific journals and can now solve advanced logical problems. PaLM 2 also has increased logical reasoning capabilities over its predecessor. This is all aimed at providing a more accurate translation. PaLM 2 is trained on over 100 languages and is trained to not only literally translate from one language to another but to apply possible context and understand idioms, poems, and riddles as well. The first improvement is a serious language boost.

This model promises to be the building block for Google's AI products, serving as a massive general LLM with expanded reasoning, language and coding capabilities. While Google did not lay out how many parameters PaLM 2 is trained on - PaLM was trained on 540 billion language parameters - Google promises that PaLM 2 will have improved capabilities in addition to faster and more efficient performance.

According to Google, PaLM 2 has already been secretly powering Google Bard and has many future integrations planned. Google is setting up PaLM 2 to be the fundamental AI model behind its future AI products.

Ai mikaze download free