Programs that detect AI-generated text can help determine whether tools like ChatGPT or Bing AI have been used.

Notable AI detection programs include GPT Zero, Detect GPT, Originality.AI

with their respective features detailed below.

However, using a detection program cannot guarantee that the text was entirely created through AI generators like ChatGPT. Even after using a detection program, it is necessary for the instructor to verify the content again.

(Example) OpenAI, the developer of ChatGPT, correctly identified 26% of AI-generated text in a test evaluation using their AI text classifier as "potentially AI-generated," while wrongly labeling 9% of human-written text as "AI-generated."

1. GPT Zero

   1) Evaluation criteria: Perplexity score / Burstiness score-perplexity change

   2)AI-generated text tends to be consistently structured, unlike human-written text, because machines generate sentences according to programmed rules. AI also creates sentences by considering which word is most likely to follow another.

case 1  Judged as written by AI(9.857/2.476)

- Lower scores indicate an increased likelihood of being AI-generated

case 2  Judged as written by a human(53.750/44.262)

- Portions of the text with a likelihood of being AI-generated are highlighted in yellow

2. Detect GPT 

   1) Evaluation criteria: Not clearly specified

   2)After entering the content and clicking ’check content’ the program judges whether the content is AI-generated and provides the likelihood

3. Originality.AI

   1) Evaluation criteria: Uses linear probability models, "Burstiness," and "Perplexity" for evaluation

   2) Human-written text is highlighted in green, while AI-generated text is highlighted in red

Source: https://raj-k-kadiyala.medium.com/gptzero-vs-chatgpt-a-gray-story-901b825e0666