© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
Google has unveiled PaLM 2, its newest and most advanced large language model that supports more than 100 languages. The large language model (LLM) is capable of performing a wide range of mathematical tests, coding, as well as creative writing and analysis.
At the core of PaLM 2 is Google’s latest TPU (Tensor Processing Units) infrastructure, which handles the computation for such models. The search engine giant says that this gives PaLM 2 an edge over its previous state-of-the-art large language models in terms of performance and efficiency. Let’s take a look at what PaLM 2 is capable of and the products it powers.
The biggest highlight of this new version of PaLM is that it’s been heavily trained on multilingual text, covering more than 100 languages. This has improved its ability to understand, generate, and translate nuanced text, including idioms, poems, and riddles across a variety of languages.
PaLM 2 also surpasses previous models in logic and common sense reasoning in mathematics. Google says that it trained the model on massive amounts of math and science texts, including scientific papers and mathematical expressions.
PaLM 2 can also interpret, generate, and debug code for over 20 programming languages. It excels at widely used languages like Python and JavaScript but can also produce code in Prolog, Fortran, and Verilog. PaLM also offers the documentation that you require in any language, assisting programmers around the world in learning coding.
Google is also heavily invested in making PaLM 2 smaller, faster and more efficient while increasing its capabilities. As such, the company has created a family of models that vary in size and performance. One of them, Gecko, is so compact that it can run as an interactive app on mobile devices without an internet connection. The other models are Otter, Bison, and Unicorn (in order of their size) are progressively larger and more powerful.
PaLM 2 has only been announced now, but it is already being used by “dozens of teams” within the company. It powers 25 products and features, including its AI chatbot Bard, which can handle multiple languages and coding tasks. It also enhances Workspace apps like Gmail, Docs, and Sheets, which leverage PaLM 2’s capabilities to assist users in writing.
That’s not all. Google has also introduced Med-PaLM 2, a specialised LLM for medical applications. It can answer questions and provide summaries from various medical texts. It also passed the US Medical Licensing Exam with an expert score. Google says it’s developing multimodal skills to process information from images such as X-rays and mammograms with hopes that this will eventually improve patient outcomes. Med-PaLM 2 will be accessible to a limited number of Cloud customers initially, later this summer, to identify and evaluate safe use cases.
Another product based on PaLM 2 is Sec-PaLM, which is designed for cybersecurity analysis. Available through Google Cloud, this LLM uses AI to analyse and explain the behaviour of potentially malicious scripts.
PaLM API has been available since March for a small group of developers, but that changes with this announcement. Google says that now developers can sign up to use the PaLM 2 model starting today.
