What can MUM do?

[kumartk]

This is now changing with the "Multitask Unified Model" (MUM), which is about 1,000 times more powerful than BERT and can even develop a kind of "understanding" of more complex information, drawing on world knowledge. This also enables MUM to conduct free dialogues on any topic, understand applied language, and even use language itself. Furthermore, the model can handle different modalities and can therefore also interpret text, images, videos, and audio information.

How does MUM work?
An example helps explain MUM's features: Users typically need more than one search query to adequately answer a complex question. This is primarily because Google provides individual pieces of information but isn't yet able to connect them.

For example, if someone asks, "I climbed Mount Adams this year and want to climb Mount Fuji next year. What should I do differently in preparation?" and wants to get information on aspects such as altitude, hiking route, temperature, and the right equipment, that person currently has to research each individual aspect and make comparisons. In the future, however, honduras phone number data Google should be able to use MUM to compare the two mountains in our example and, among other things, highlight that they have roughly the same altitude. In addition, the search engine could point to suitable equipment and, through MUM, access text, image, audio, and video content from across the internet – completely independent of the source language.

Furthermore, its multimodality will allow MUM to accept search queries in a variety of ways: For example, users could submit a picture of their hiking boots to Google and ask if they're suitable for climbing Mount Fuji. A possible response could then be a link to a blog about suitable gear for climbing that mountain.


Figure 1: “Google, can I climb Mount Fuji with these shoes?” – Source: Google
The meaning of entities
Over the years, Google has improved not only its search algorithm but also the way it stores information and displays it in the SERPs. In 2012, with the Knowledge Graph update and the slogan "Things, not Strings," Google addressed the importance of entities for search, stating that the search engine can actually interpret the meanings behind words and provide relevant answers based on previous searches and related content.

An entity is a uniquely identifiable unit about which information can be stored. Almost anything can be an entity, including people, objects, or cities. For example, a well-known person can represent a separate entity, for which attributes such as age and place of birth are stored. Additionally, connections can be made to other entities related to that person. Because Google can recognize the different nodes between the individual entities, a vast, dynamic information network is created that can be accessed at any time.