The Google company has been behind numerous stages of AI development around the world and has made many breakthroughs in recent years. However, he did not have time to complete his own language model and Bard chatbot and lost the lead to ChatGPT. In a panic, the company had to launch its AI early, which led to problems in its work and subsequently harsh criticism from both users and the company’s own employees. The development of OpenAI ushered in a new era of artificial intelligence that sets strict rules: from now on, anything that can be connected to neural networks must be connected to them. That’s why Google tries to integrate development into the maximum number of its services.

What is known about PaLM 2:

Google DeepMind Vice President Zubin Ghahramani and other Google executives introduced PaLM 2 on Wednesday, May 10. The major language model is now available in four scales and can run natively on any device, from smartphones to entire server farms.

PaLM 2 is apparently the successor to Google’s existing PaLM model, which until recently was the basis for Bard’s experimental AI.

Think of PaLM as a general model that can then be fine-tuned to perform specific tasks. For example, health research groups have developed PaLM with medical information in mind to help answer questions and summarize information from various medical texts.
Gahramani explained.

• The company said Bard is already working on PaLM 2, which offers advanced multilingual, reasoning and coding capabilities. The language model was much better trained on multilingual texts than the previous model; Covers 100+ languages ​​with advanced understanding of cultural idioms and expressions.
• It produces Python and JavaScript code equally well.
• The model also reportedly demonstrated “advanced abilities in logic, common sense reasoning, and mathematics” thanks to large amounts of training data from “web pages containing scientific papers and mathematical expressions.”

Even more impressively, Google has been able to release application-specific versions of the core PaLM system called Gecko, Otter, Bison, and Unicorn.

We built PaLM to be smaller, faster and more efficient while expanding its capabilities. Next, we created a family of models in a wide variety of sizes so that the lightest model can work as an interactive app on mobile devices based on the latest Samsung Galaxy model.
– says Gahramani.

In total, at the I/O event, Google announced more than two dozen products that will feature PaLM capabilities. We’ll talk about them later.