Yesterday, during its annual developer conference (I/O 2023), Google announced the (PaLM 2) model, the latest large language model (LLM) it has developed and the model in which the chatbot (Bard) operates, and it will also be the cornerstone of all the artificial intelligence products that Google offers thanks to its extensive capabilities and possibilities.
PaLM 2 is the first and direct competitor to the (GPT-4) model introduced by (OpenAI) last March and currently running the paid version of ChatGPT.
Zoubin Ghahramani, vice president of Google company DeepMind, said in the post announcing the model on the Google blog: “Over the last decade we have learned a lot from the development of artificial intelligence, such as how we can intelligently scale neural networks, how we do it.” Learned during our research. Model size is not the main component of language models. Bigger isn’t always better. Research creativity is key to building great models. Recent advances have taught us how to design and train models, how to embed multimedia, the importance of human feedback, and how to build models more efficiently than ever before.”
Here’s everything you need to know about Google’s new PaLM 2 AI model:
First; What is the PaLM 2 form?
Related topics to what you are reading right now:
(PaLM 2) – the new generation of the Language Large Model (PaLM) first introduced by Google in 2022 – is a language model that is faster and more efficient than the previous version thanks to its strong abilities in logical thinking and mathematics, writing and debugging code and deal with multilingual texts.
According to Google, the (PaLM 2) model has been extensively trained on multilingual text spanning more than 100 languages and this has greatly improved its ability to understand, generate and translate concise text with greater accuracy than previous models.
Google also focused on improving its language model in the areas of reasoning and mathematics, as (Qahramani) pointed out that the company applied the model to a large body of mathematical and scientific texts in addition to scientific papers and websites containing mathematical expressions had trained.
It’s no secret that large language models, with their focus on language, have had difficulty answering mathematical questions without resorting to external tools. According to Google, its model (PaLM 2) can easily solve math puzzles, solve problems in a logical way, and even render charts without the need for external tools.
The PaLM 2 model thus forms the basis for Google’s Codey model for code generation and debugging, which Google also announced yesterday as part of its code completion and generation service.
Second, Google offers a variety of templates:
The PaLM 2 model is more powerful than previous models and also faster and more efficient. Google offers it in a variety of sizes, making it easy to rely on for a variety of use cases.
Google says it will offer the PaLM 2 model in four different sizes, from smallest to largest: gecko, otter, bison, and unicorn.
The Gecko model is lightweight enough to work on mobile devices and is fast enough to work with interactive mobile applications even when offline. This versatility means PaLM 2 can be tuned to support entire product categories in more diverse ways to help more people.
Third; For how many products does Google use the PaLM 2 model?
Google announced more than 25 new products and features based on its work on the (PaLM 2) model yesterday at the (I/O 2023) conference. This means that this model can offer the latest advanced artificial intelligence capabilities directly to people including : consumers, developers and businesses of all sizes around the world.
Here are some examples of Google products using the PaLM 2 model:
1- Chatbot (cool)
The improved features of the PaLM 2 model allowed Google to extend the capabilities of the Bard chatbot to new languages. The robot now supports three languages: English, Korean and Japanese and hopes to expand the list of supported languages to include 40 more languages in the coming period. In addition, code generation and debugging are now supported.
2- Features of artificial intelligence in workspace applications:
The AI capabilities Google announced in its Workspace suite — which help generate text from simple prompts in Gmail messages and Google Docs, and organize spreadsheets — are based on the PaLM 2 model and help people do their jobs more efficiently, to do better and faster.
3- Med-PaLM 2 Medical Knowledge Model:
Google has also announced a model (Med-PaLM 2) that focuses on medical knowledge as it can accurately answer medical questions and summarize ideas from a variety of medical texts.
This model is the first major language model operating at the expert level on medical issues; Accuracy in answering United States Medical Licensing Examination (USMLE) questions is greater than 85%. Google is now adding multimedia processing capabilities to aggregate information like X-rays and mammograms and one day improve patient outcomes.
Google has announced that it will make the Med-PaLM 2 model available to a small group of its cloud service customers later this summer to gather feedback to identify safe and useful use cases.
4-Sec PaLM Electronic Security Model:
Also yesterday, Google announced Sec-PaLM, a special version of PaLM 2 aimed at security use cases and available through Google Cloud.
The Sec-PaLM model uses artificial intelligence to analyze and explain the behavior of potentially malicious scripts and will, in an unprecedented time, be able to detect scripts that actually pose a threat to individuals and organizations.
5- Availability of the PaLM model for developers:
Since March, Google has been testing the PaLM API with a small group of developers, and since yesterday developers can register to use the PaLM 2 model. Alternatively, customers can leverage the model via the Vertex AI platform for enterprise-level privacy security. .
Developers and enterprises are already using the Vertex AI Platform on Google Cloud to build and deploy machine learning and AI application models at scale.
Fourth; What plans does Google have to advance the future of artificial intelligence:
Google said yesterday that (PaLM 2) demonstrates the impact of powerful models of different sizes and speeds, and that versatile AI models offer real benefits for everyone, and is therefore committed to releasing the most useful and responsible AI tools.
She added, “The teams at Brain and DeepMind have achieved many pivotal moments in artificial intelligence over the past decade, and today we are uniting these two world-class teams as one to continue to accelerate our progress. not just today, but responsibly.” paving the way for the next generation of AI models.”
In addition, Google is working on a new model called (Gemini), which should be multimedia and highly efficient in integrating tools and APIs, and should allow for future innovations such as long memory and layout.
The (Gemini) model is currently still in the training phase, but Google has stated that it already offers enormous multimedia handling capabilities that were unprecedented in previous models, and Gemini will also be available in different sizes and capabilities, exactly like the ( PaLM 2) model, to ensure that the ability to use it across different products, apps and devices benefits everyone.