This seems pretty useless, because won’t GPT3 happily synthesise false information? What good is it answering my question if the answer isn’t correct? Google search’s “question answer” section has a similar issue.
You can condition the prompt to make it say 'I don't know' etc. which is what we actually did too. Give it a try with a random question.
P.S. This is meant to be integrated to individual vendors' websites so the data we will get from will be much more detailed, making answers much more to the point
P.S. This is meant to be integrated to individual vendors' websites so the data we will get from will be much more detailed, making answers much more to the point