While CAPTCHAs have been around for years as a way to differentiate between human and AI traffic, their effectiveness has decreased, and many consider them largely ineffective.
Everyday users will be familiar with encountering a CAPTCHA that seems impossible to solve, and often these tests are unsuccessful, leading to retries or the use of an accessibility pass (which could be exploited).
Some have posited that the rise of AI will render CAPTCHAs obsolete, and AI company Cohere has run experiments to test the limits of modern CAPTCHAs with its large language model ChatGPT.
In a series of tests, ChatGPT was presented with various CAPTCHA types, including identifying numbers drawn in front of scenes, finding similar images in a selection, and selecting the image that did not fit a set theme.
The large language model performed well on the straightforward tests, but when faced with images that were distorted, generated by AI, or that relied on detecting similarities, its performance suffered.
