Skip to content
AImpact
IT EN
Medium Multimodal AI · 1 min read

MUM: Google unveils the multitask model for Search

In one sentence At Google I/O, Google announces MUM (Multitask Unified Model), built on the T5 framework, trained across 75 languages, multimodal (text and images), and, per Google, 1,000 times more powerful than BERT.

Verified Official source
ShareLinkedInX
Reading level

During Google I/O, Google announces MUM (Multitask Unified Model), a new model meant to better answer complex questions in Search.

The example shown: you've already hiked Mt. Adams and want to hike Mt. Fuji next fall; what should you do differently to prepare? Today it takes many separate searches on elevation, weather, trail difficulty, and gear (Google says complex tasks take eight queries on average). MUM should understand the question and assemble the answer, for instance pointing out that fall is Fuji's rainy season and a waterproof jacket may be needed.

For now it's an announcement: Google says MUM-powered features will reach its products in the coming months and years.

Companies

Google

Tools

MUM

Tags

GoogleMUMSearchT5Multitask

Sources