Urgent.News

What's breaking now, across thousands of outlets.

AI

Helping AI text detectors keep up with changing conditions

A more flexible training method can make AI text detectors more robust when new models emerge or AI-generated text is disguised to appear human.

Helping AI text detectors keep up with changing conditions

A new paper from MIT and Cornell University researchers reveals how AI text detection tools can be deceived by AI-generated text and how to make them more accurate. While teachers, editors and other users may rely on AI detection tools to confirm whether a term paper is written by a human or an LLM, these tools can be fooled by text that appears to be written by a human.

Additionally, as writing habits change over time and new LLMs are released, AI detectors can fail to identify AI-generated text. To tackle this issue, the researchers demonstrate a technique called test-time adaptation (TTA), which enables an AI detector to adapt to new, unlabeled data on the fly, making it more accurate and resilient to deception.

In experiments, a commercial detector could only identify about 27% of AI-generated text that had been disguised, while TTA achieved 93% accuracy. This method could be used by educators, editors, publishers, and hiring managers to flag LLM-generated writing that may be presented as human-written, and by cybersecurity experts to prevent phishing scams or identify search engine optimization fraud.

Written by urgent.news from MIT News AI's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Other reports of this story

Every outlet that reported this, and when →

Read the original at news.mit.edu →

More in AI

More from Wednesday 18 November →