Google announced its next big change in how it understands queries and language on the web – they are calling it MUM – Multitask Unified Model. This is not live yet but Google wanted to share a bit about it. Google said this “new technology Google is exploring internally to better understand language & make it easier to get helpful responses to complex Search needs.”
Google said it is like BERT in that it built on a Transformer architecture but it is 1,000 times more powerful. MUM not only understands language, Google said but it also generates language. Google said MUM is “trained across 75 different languages and many different tasks at once, allowing it to develop a more comprehensive understanding of information and world knowledge than previous models. And MUM is multimodal, so it understands information across text and images and, in the future, can expand to more modalities like video and audio.”
Here is the example:
In the example above, Google said MUM can surface insights based on its deep knowledge of the world, it could highlight that while both mountains are roughly the same elevation, fall is the rainy season on Mt. Fuji so you might need a waterproof jacket. MUM could also surface helpful subtopics for deeper exploration — like the top-rated gear or best training exercises — with pointers to helpful articles, videos and images from across the web.
It is also really good with languages and understanding information across types. Google said “MUM-powered features and improvements to our products in the coming months and years.”
Glenn Gabe shared a paper on this:
A new Google -> “A team of Google researchers published a proposal for a radical redesign (of Search) that throws out the ranking approach & replaces it with a single large AI language model, such as BERT or GPT-3—or a future version of them.” https://t.co/AcBoEB6Q3O pic.twitter.com/bB0P9aPuKZ
— Glenn Gabe (@glenngabe) May 17, 2021
You can learn more at the Google blog.
Forum discussion at Twitter.