AI Writing
Do Humanize AI GPTs Work To Bypass AI Detectors?
Explore the world of specialized GPT models in ChatGPT's GPT Store. Discover humanizer models and delve into their impact on AI content detection.
With the launch of ChatGPT’s GPT Store, users can now access GPT models that are fine-tuned for specific types of assistance. Ranging from creative writing to legal advice, technical coding and even medical information, these specialized GPTs go well beyond the ordinary ChatGPT model.
When it comes to tools attempting to bypass AI detection GPTs that claim to bypass detection do not work as well as dedicated undetectable AI writers.
But no undetectable AI writer or adversarial technique reliably bypasses Originality as shown in published study.
Since the launch of the GPT Store and not surprisingly, some of the most popular GPTs are “humanizers”. Humanizers are designed to subvert the robotic “AI-sounding” GPT writing and replace it with content that sounds more human.
But are these “human-style” models really able to create undetectable AI content?
We put them to the test.
We chose the most popular GPTs from the Writing section of the GPT store, including Humanize AI and Humanizer Pro.
For this quick test, we took five known AI samples from our AI detection benchmark dataset. We then ran the text through Humanize AI and let Originality.ai check it for authenticity. We then repeated these steps with Humanizer Pro.
For those of us who work every day to uncover AI-generated writing, the results were hardly surprising.
As a baseline measurement, we ran the default AI detection test through Originality.ai. It guessed that all five samples were correctly AI-generated, as expected:
ChatGPT Generated Text:
We then took the original AI-generated samples and ran them through Humanize AI from the GPT Store. After the text was “humanized”, we fed it back into Originality.ai with these AI detection results:
Humanize AI Modified Text:
We repeated the same steps above but for the second test, we ran the AI-generated text through Humanizer Pro. After being humanized by that particular GPT, we then fed the text to Originality.ai with these results:
Humanizer Pro Modified Text:
See raw data here.
It’s understandable to want some kind of shortcut that will differentiate AI-generated writing and make it sound more human. Unfortunately, our tests proved that even with these specialized GPTs, AI still has a long way to go before it can sound authentically human-like. As new iterations of these “humanizing” GPTs are released, we’ll repeat and update our tests accordingly.
However, as of right now, even specialized GPTs from the GPT store that promise to “humanify” your text will fall short when put through their paces with true AI detection software. If you’re thinking of using a humanizing GPT to make content undetectable, think twice, as the best AI writing detectors can still pick up on it with nearly 100% accuracy each and every time.
