Ghostbuster vs. ChatGPT: Can You Detect AI Generated Text?

Instances of cheating in exams and interviews, fabricating research, articles and news, and other forms of dishonesty through AI-generated text are on the rise, and there were many attempts to detect LLM, mostly ChatGPT generated text.

Join our next cohort: Full-stack GenAI SaaS Product in 4 weeks!

While tools like DetectGPT and GPTZero have made commendable efforts in this domain, they often face hurdles such as deviation from training data and false positives, leading to potentially unfair outcomes.

In response to these challenges, a team of researchers from Berkeley AI Research (BAIR) has developed Ghostbuster, an advanced tool designed for the detection of AI-generated text.

Ghostbuster’s methodology has proven to be significantly more effective than its predecessors in most testing scenarios:

Results of our model across a variety of text domains (F1): https://arxiv.org/pdf/2305.15047.pdf

In this article, I will walk you through:

• How the team generated training data for such a task
• Model training process
• Setting up Ghostbuster locally
• Detecting AI-generated text with Ghostbuster
• The code that generates scores

Let’s find out who’s cheating!