Let’s talk about an advancement in Bard, Google’s answer to ChatGPT, and how it addresses one of the most pressing problems with today’s chatbots: making stuff up. Last year, their makers warned us not to trust chatbots. Tools like ChatGPT generate text without drawing on a database of known facts. Chatbots instead predict words based on their underlying large language models that have been trained on a massive corpus of text.
Google has now given Bard a new way to check if its answers are correct. If you ask Bard a question and it gives an answer, you can click the “G” button to get Bard to check itself. Bard will read its own response and see if it can find information online to prove the answer right or wrong.
When you click the “G” button, words in green mean Bard found websites backing up that part of the answer. Words in brown mean Bard didn’t find anything proving that bit. This helps you know which parts of the answer may be wrong.
Bard double-checked an answer about the band Radiohead for example. It highlighted most of the responses in green since it found websites agreeing with those facts. But it put one sentence in brown saying they won nine Brit Awards. By looking closer, Bard saw they never won that many.
Even with this new way to check itself, Bard isn’t perfect yet. But it’s a good step to help chatbots admit when they may have the wrong information. In the future, researchers hope the bots can check their work better without needing human help as much. This update makes Bard a little more trustworthy when looking for facts online.
