Tuesday, May 30, 2023

Asking Generative AI to Cite Sources - Which Tool Is Best?

Image of a diverse range of books, documents, or articles with a magnifying glass hovering over them, symbolizing the concept of information verification and citing sources.

In the
"6 Tenets of Post-Plagiarism: Writing in the Age of Artificial Intelligence" Dr. Sarah Eaton explains that using generative AI in writing will become normal. The writing process has now shifted, and while AI may do some, or even most, of the writing, humans are responsible for fact-checking, verification, and knowing the source of the information as applicable. 

Popular generative AI tools include Bing, Google's Bard and Help Me Write, and ChatGPT. While you can get sources from each, they are not equally good at easily providing accurate sources. Let’s take a look.

Comparison of Generative AI Model’s Ability to Cite Sources

Google

Google’s Bard and Help Me Write, do a mediocre to poor job of citing sources. 


When asking Bard to cite sources, it is not unusual to get a response such as: 


I'm a text-based AI, and that is outside of my capabilities.

Or…

I can't assist you with that, as I'm only a language model and don't have the capacity to understand and respond.


With some additional prompting, you can often get past that, but even when it cites sources, it often may cite an indirect source. For example, instead of the direct source, it seems to generally cite a source such as Wikipedia. 

ChatGPT

Citing sources, is usually an afterthought with ChatGPT. Unlike Bing, the sources are generally not embedded. Even when prompted, ChatGPT may list some sources, but they may or may not include links directly to that source, and they are often not embedded in results. Sometimes they are inaccurate, so you must check and verify them.  

Bing

Bing is currently the best generative AI tool when it comes to citing sources. It will cite sources without needing additional prompts and the sources are generally accurate and direct. 


Let's take a look at the elements of each as it relates to source citation.

Comparison Chart 



ChatGPT-4

Google's Bard / Help Me Write

Bing

Real-Time Updates

No

Sometimes

Often

Strengths in Citing Sources

Can be guided to include citation-like structures, but does not inherently cite sources

Can be guided to include citation-like structures in narratives, but does not inherently cite sources. Even when guided, the sources are often not direct.

Will usually provide accurate links to source material.

Currency

Currently only has data up until September 2021 so information may not be the most up-to-date and accurate.

Not versatile in citing sources, though it does have the ability to access up-to-date information.

Access to real-time information. Robust at searching up-to-date information.

Summary

Provides the best content, but does not provide up-to-date and accurate source citation.

Provides mediocre content and only sometimes provides reliable and useful sources when prompted.

Usually embeds accurate, up-to-date and reliable sources with direct links.

The Verdict

While Google and ChatGPT do a poor job of citing sources, each does have advantages. ChatGPT will provide the most robust and comprehensive response. Google’s Help Me Write is nice, because it’s embedded directly into Google Docs, so you can just ask for help while in a document rather than having to go elsewhere. 


However, when it comes to citing sources, Bing is the winner. However, the output is not as strong as ChatGPT, so it often makes sense to use a combination of Bing and ChatGPT for now. 

Saturday, May 6, 2023

Embracing Post-Plagiarism: What It Means in the Age of ChatGPT

The rise of artificial intelligence (AI) has had a profound impact on education. As AI-powered tools continue to evolve, they are transforming the way we approach learning and teaching. Dr. Sarah Eaton published an insightful article titled "6 Tenets of Post-Plagiarism: Writing in the Age of Artificial Intelligence" on her blog. Take a look at the post for the full explanation of each tenant. Each provides a simple, yet powerful insight. 

Two powerful tenents to keep in mind are as follows:


Hybrid Human-AI Writing Will Become Normal

We've been using predictive text, spell check, grammar check, extensions, and more for at least the past decade. Those who have the ability can also ask family, tutors, and others for feedback. AI does this on steroids and can dramatically help an author improve their writing. As Eaton says, "Trying to determine where the human ends and where the AI begins is pointless and futile." Good writing usually incorporates feedback and assistance and there is nothing wrong with that. We can invite and even relinquish control to others, but her next tenant is key. 

Humans can Relinquish Control, but not Responsibility

Humans retain control over what they write, but the best writers usually ask for input from others. These days, that may include AI. These writers understand there is always room for improvement and asking for feedback to improve writing is a strength. However, when asking for feedback, it is up to the author to be accountable for fact-checking, verification, truth-testing, and ensuring their voice comes through. Humans are also responsible for ensuring that AI tools are ethically developed and non-bias. 

Read all six tenants on the blog post and/or in the graphic below.  

Visit the original post for the text of this graphic, video, and more details.