Finding relevant references for information generated by ChatBots (like ChatGPT) using Google Bard

Since the launch of ChatGPT back in November 2022 by OpenAI, every other company have also jumped in this race of developing & then deploying AI based chatbots with which people can have conversations.

Although on first sight of using these chatbots, it seems like magic.

Just ask it something and then boom it will generate a response.

But that response may contain information which is not true at all, chatbot just made up that information and showed it, that’s called Hallucination. It’s doing so because of way it works, its been trained on finding patterns & relationships between words, phrases, sentences by providing it large amount of text (or video/audio in some cases) which is available out there on Internet (webpages, books, research papers etc.)

Whenever your asking it something, then its just predicting next words in sentences and it have no idea - if overall sentence that its generating is actually true or not (although it can be fine tuned to a certain extent for generating more accurate responses but achieving 100% accuracy of now in 2023 is not possible at all)

As ChatGPT don’t have any way to either verify or find some relevant references to what its showing in response to your question, that’s why its really hard to fully trust it.

But another chatbot Google Bard (its a Google product which they launched as competitor to ChatGPT). Both of ChatGPT and Google works using similar tech that ChatGPT uses - Transformers Architecture as of now in Sep 2023, only major difference is that Bard have access to internet so if you ask it something recent, it can answer your question while ChatGPT can’t.

Recently just couple of days ago on 19 September, Google launched some new Bard features and one of them is double-check Bard’s responses with Google and in their own words they described this feature as following.

What: With the power of Google Search, the [G] button can help you double-check Bard’s AI-generated responses. Starting with English, when a statement can be evaluated, you’ll see it highlighted in Bard’s responses and can click to learn more.

Note: The links provided are content found by Search and do not imply these sources informed Bard’s initial response.

Why: People are using AI tools to more easily understand complex topics in new ways. As you continue your learning journeys with Bard, it’s important that you feel more confident in the information generated by AI.

So essentially what Google did is to implement a feature which can be used for verifying information contained in a response which Bard generate.

For example - If I ask Google Bard “How to get good in Python Programming?“ then it will generate a response and then if I click on little google icon at bottom of its response then it will take couple of seconds. After which some lines will be highlighted with orange kind of background color while others with green (also its not like it will highlight every line in response, only couple of them will be highlighted)

Green highlighting means that Bard found something related to that specific line on a webpage and then you click on see what that webpage is. Then if you click on it, it will either take you to that webpage or sometimes even highlight specific paragraphs on that page. While orange highlighting just means that Bard tried to look for some related information to that specific piece of text, but it did not found anything.

So Google Bard can be used for generating some web references for text which it generates for question responses, but can somehow it be used for finding some relevant web references for text which is generated by other chatbots like ChatGPT. I’ve been interested in information generated by ChatGPT since like March because I use it extensively almost every other day to help me understand different concepts mentioned in research papers.

That’s why as soon as I heard of this verify response announcement by Google Bard, I jumped on to figure out to find some way to find relevant references for text generated by other chatbots and after spending around 3 & half hours yesterday on Sunday afternoon, I finally figured it out ??

Here’s how you can use Google Bard to find out references for text generated by other chatbots like ChatGPT.

Ask Google Bard following If I provide you some paragraphs, can you just respond back to me with those paragraphs? Nothing added or removed, just respond back whatever paragraphs I provide.

Then ask Google Bard following

Ok here are paragraphs (provide some paragraphs here)

Once you done with that then Google Bard will respond back to you with whatever paragraphs you have provide it already, then just scroll down to end of it.

Then just click on that little Google icon, then after just around a min or so. Bard will highlight some lines with either green or orange background.

Green will be references to webpages while orange shows that although Bard tried to look for similar information to that text but it didn’t found anything related on Internet.

Boom!!! That’s how you can use Google Bard to even find references for text which is generated by other chatbots like ChatGPT.