Refresh
One of Bard’s interesting quirks – and a point of difference from its chatbot rivals – is that it’ll present you with a few drafts of potential responses, showing that it’s very much a tool for creation rather than fact-finding.
This is known as ‘AI branching’ and we think it’s a handy tool. We’ve already seen that these AI-generated drafts are coming to the likes of Gmail and Google Docs soon too, alongside a host of other handy cheat codes (sorry, features) like automated presentations and smarter auto-fill functions for Google Sheets.
So how exactly was Google Bard trained? Google has already explained that it’s based on a lightweight version of its large language model called LaMDA, but we wanted to get Bard to explain its own origin story.
There were no bites from radioactive spiders involved, sadly – just the usual tale of being locked in a room with an unimaginable number of “books, articles, websites and code”.
Supervised learning was involved and Bard says it took the “long time” of “several months” for it to learn how to communicate like a human. Wait till it hears how long it takes us humans.