Introducing Pangram’s Side-by-Side Analysis

If you’re a teacher, assigning an essay or other writing assignment to your students is no simple task.

You put thought, care, and time in to clearly articulate to your students what you’d like them to write. You carefully design assignments to spark new connections, promote deeper reflections, and challenge your students to think bigger.

All that care, and with new AI models, it can take under 15 seconds for a student to submit your assignment description to ChatGPT and receive a nearly perfect essay back. No care, no time, and definitely no learning.

Oftentimes, when you get these essays back, you can see the signs of AI use. Maybe it’s the common AI phrases, or a usual type of formatting. You know that if you put your assignment into ChatGPT, you could get a very similar essay.

But then what? You could deep read the text, highlight any similarities, and present your suspicions? Doing this for every essay you suspect of AI use and compiling all of this evidence together could take hours.

Now, Pangram can do that for you. Here’s how:

The Solution

First, put an essay you believe to be AI-generated through the dashboard at pangram.com

If it is indeed AI, it will show up as follows:

Pangram Dashboard - AI Detected

Next, scroll down to the bottom of the page to view “Compare with AI”. Click on that panel to open it!

Side by Side Analysis Panel

In this text box, you should paste your assignment instructions. Put anything you believe may have been relevant to the student when they were creating their essay or writing assignment with AI. Then, click submit!

Generated Side by Side Analysis

Within 10-15 seconds, you should see an interface like above! Let’s break down exactly what’s going on here.

First, Pangram prompts the LLM we believe generated the essay with the text of your assignment, exactly how a student who may have pasted your instructions into ChatGPT would have done.

Next, we run an algorithm over both texts to extract semantically similar phrases. This means that we will extract exact matches of text from both essays, but also phrases that are similar in meaning. Let’s take a look at this Great Gatsby example.

As you can see, this is a combination of exact matches and similar phrases. In this essay alone, there are many semantic matches, each of which can be used to bolster an academic integrity case or similar evidence-based argument for why a particular document is likely AI-generated.

This new tool is part of Pangram’s continuous effort towards adding more explainable features to back our industry-leading AI detection model. Interested in learning more? Feel free to reach out at info@pangram.com!