With announcements at its developer conference I/O, Google has once again fueled the exchange of blows with rival Microsoft and ChatGPT developer Open AI, which has been going on for months. It will be a "busy year for artificial intelligence," said Google CEO Sundar Pichai, presenting the company's new major language model on stage for the first time: PaLM 2.
Editor in the economy
- Follow I follow
Such language models are the basis for applications such as ChatGPT. They are trained with vast amounts of data and are therefore able to answer even complex questions at a high level. The new model should be able to do this faster, better and more efficiently, Pichai promised.
In addition, it has new capabilities compared to its predecessor: While the current models were mostly fed primarily with English data, PaLM 2 is explicitly trained on more than 100 languages at a high level, according to Google. In this way, artificial intelligence can understand the nuances of individual languages, for example formulate poems and riddles. A developer explains this with an example: The German phrase "I only understand the station" would have taken previous models literally. PaLM 2 would understand the actual meaning.
Because the model was specifically trained with mathematical and other scientific content, the program also has better logic capabilities. The programming capabilities of PaLM 2 are also on a new level. The model is even smaller than its predecessor in terms of the so-called parameters, explains a developer. Parameters are the values that the AI learns during its training and refines through further training. The models have become larger at a rapid pace in recent years. The fact that Google is now moving in the other direction again is a turnaround.
25 new AI-based applications
PaLM 2 is available in four different sizes. According to Google, the smallest model even runs on smartphones. Which of the four sizes fits best depends on the application. The more specific the requirements, the smaller the model can be. The more general the scope becomes, the larger the model needs to be, because it needs more context.
Google announced 25 new products and applications based on PaLM 2 at I/O. This includes Google's own AI chatbot Bard, which is finally set to pick up speed after a difficult start. Google Bard has recently secretly switched from the previous LaMDA model to PaLM 2. Google also announced that it would make Bard freely available in more than 180 countries. Previously, only select users in the United States and the United Kingdom were able to access the program. Bard now also understands Japanese and Korean, with 40 more languages to follow quickly with the help of PaLM 2. And Bard should also soon be able to process images. Answers from Bard can now be exported directly to Gmail or Google Docs.
Watermarks for AI images
Other applications such as Gmail and Google Maps are also to receive new functions based on artificial intelligence. Pichai demonstrated, for example, how a program can automatically write an e-mail to refund airline tickets. New image editing software is designed to allow users to use artificial intelligence to add parts of the image that are not even visible in the original photo.