The company, typically secretive about its algorithm, shared the changes at a media briefing. Its search function can now handle more complex inquiries through a new sequence of code called Bidirectional Encoder Representations from Transformers—or Bert—that uses advanced machine learning. It’ll improve one in 10 queries, Google said. (Wall Street Journal)

Talking point: Google executives claim the changes are among the biggest improvements to search in five years. Previously, queries with slang or words that have multiple meanings would confuse the search engine, which would interpret them literally or bring up irrelevant results. Bert is intended to overcome those issues, partially by ignoring unnecessary words like transitions. Google’s search capabilities are at the centre of its digital ad business, which makes US$136 billion a year. The company doesn’t share the exact number of searches it receives per day, but some estimates place the figure at 3.5 billion. That means Bert could be improving some 350 million queries a day.