Google developers in their blog talked about a number of innovations in the search engine. One of them is the improvement of snippets — fragments of text that are displayed above the issue and provide answers to specific user questions.
Previously, the search engine could catch on to one part of the question, not considering the other. As a result, the user could get an answer to an absolutely absurd question. For example, if you ask “When did Snoopy kill Abraham Lincoln?”, then Google, as if agreeing with this fact, displayed the exact date.
Now new artificial intelligence models help the system understand when the piece of text it has selected does not fully match the user’s question. As a result, issuance in such cases improves by at least 40%.
Also with the help of the new Multitask Unified Model (MUM), the search engine now uses the concept of consensus, where multiple high-quality sources on the Internet agree on the same fact. This increases its weight and allows you to improve the snippet.
Another improvement is a warning in the issue that the search engine is not sure about the quality of the suggested answers. This is relevant for situations where people are looking for data on some kind of breaking news, where events are developing quite quickly.
Read also 🧐