Putting Undetectable AI Writers to the Test: Can They Fool AI Detectors?

Table of Contents

• Introduction
• The AI Writers and Detectors
• The Tests
• The Results
• The Winners
• Conclusion

Introduction to AI Writers and Detectors

AI writers have advanced rapidly in recent years, with some touting the ability to generate human-like content indistinguishable from text written by a person. This has sparked interest in AI detectors designed to analyze text and determine if it was AI-generated.

We conducted an experiment putting leading 'undetectable' AI writers up against top AI detectors. The goal was to see if AI-generated content could truly fool the detectors designed to catch it.

Examining the AI Writers and Detectors

For our test, we selected 10 top undetectable AI writers that claim to produce human-like content: GPT-INF, Hide My AI, Stealth Rider, Undetectable AI, Stealth GPT, One-Click Human AI, Text Converter, Right Human, Humanized AI, and Undetectable GPT.

We pitted them against 20 leading AI detectors: Winston AI, GPT-0, Check for AI, Originality, Interested in AI, AI Cheat Check, AI Undetect, Content Detector, Content at Scale, AI Checker Tool, Zero GPT, AI Text Detector, SEO AI Content Detector, Checker AI, GPT Kit, GLTR, Content Checker, Detech GPT, and GPT Safe.

We invested $269 to purchase credits for all the AI writers and detectors to thoroughly test their capabilities.

How the AI Writers Work

The undetectable AI writers function similarly, providing an interface to paste text for rewriting. You select options like 'paraphrase' or 'humanize' and the AI attempts to rewrite the original text to be undetectable. The core premise is that they're trained on vast datasets of human writing, allowing them to mimic patterns and quirks inherent to how people write.

How the AI Detectors Work

The AI detectors also tend to provide a simple interface for pasting text excerpts to scan. They analyze lexical patterns, semantic consistency, grammar, and other linguistic factors to estimate the likelihood text was AI-generated. Some claim near-perfect accuracy scores based on testing against known AI content. The goal is developing models that can flag machine-generated text reliably.

Testing the AI Writers Against Detectors

We used ChatGPT to produce unique samples of text for rewriting: a 500-word essay on colonial America, a 500-word software engineer cover letter, and a 500-word fantasy short story.

This provided sufficient content for the AI writers to attempt rewriting and for the detectors to analyze the output.

Test 1: Colonial America Essay

In the first test focused on the colonial America essay, the AI writer Hide My AI performed best, fooling all detectors except Winston AI and AI Cheat Check. Undetectable AI and Stealth Rider took second place, each being caught by those same top two detectors. Humanized AI unfortunately scored last, being unable to convincingly rewrite the essay.

Test 2: Software Engineer Cover Letter

On the second test using the software engineer cover letter, Hide My AI and Stealth Rider tied for first place this time. They slipped past all detectors except for Winston AI and AI Undetect. Stealth GPT took second place, while Humanized AI again scored lowest.

Test 3: Fantasy Short Story

For the final fantasy short story test, GPT-INF scored a perfect result, fooling 100% of the detectors. Hide My AI took a very close second, being caught by just one detector. AI Text Converter scored lowest this round, but still passed most detectors.

Evaluating the AI Writer and Detector Results

Looking at average performance across all test cases, Humanized AI unfortunately scored as the least convincing 'undetectable' AI writer. It seems improvements are still needed so it can mimic human writing patterns more accurately.

Hide My AI edged out the competition as the most consistently undetectable AI writer. Interestingly though, GPT-INF scored one perfect round, demonstrating that AI generation tech continues advancing rapidly.

The Top Undetectable AI Writers

Considering both overall consistency and perfect scores, the winners for most undetectable AI writers are:

• Hide My AI - $5/month for paid version
• GPT-INF - $12-$349/month for paid tiers

The Top AI Detectors

On the detector side, Winston AI dominated with near-perfect accuracy, being fooled just twice overall. GPT-0 and Check for AI tied for second place. So the clear winner is:

• Winston AI - $112-$192/month for paid tiers

Key Takeaways and Next Steps

This experiment yielded fascinating results showing leading undetectable AI writers can indeed bypass many detectors, yet the top detectors remain difficult to fool.

As both sides continue progressing, an AI-generation arms race is unfolding. We plan additional head-to-head tests evaluating iterations of these tools over time.

If interested in evaluating AI writers and detectors yourself, please check the pricing details mentioned previously. The tech remains immature but advances rapidly.

We welcome you to subscribe for future videos assessing advancements in this emerging domain of undetectable AI generation. Thank you for reading!

FAQ

Q: What was the purpose of this experiment?
A: To test how well the latest 'undetectable' AI writers can evade detection from AI content detectors.

Q: Which AI writers and detectors were tested?
A: 10 undetectable AI writers were tested against 20 AI content detectors.

Q: What kind of content was tested?
A: The AI writers were given 3 prompts to rewrite: a colonial America essay, a software engineer cover letter, and a fantasy short story.

Q: Which AI writer performed the best?
A: Hide my AI and GPT INF were the top performers, with the fewest instances of detection.

Q: Which AI detector performed the best?
A: Winston AI correctly identified the most AI-generated content.

Q: Did any AI writer achieve a perfect score?
A: Yes, GPT INF's fantasy short story rewrite completely evaded detection.

Q: Did any AI detector catch all the AI content?
A: No, even the best detectors were fooled some of the time.

Q: What was concluded from the experiment?
A: The top AI writers can evade some detectors, but the top detectors still catch most AI content.

Q: How can I get an undetectable AI writer?
A: Two top options are Hide My AI ($5/month) and GPT INF (starting at $12/month).

Q: Where can I get a good AI content detector?
A: Winston AI emerged as the top detector (starting at $19/month).