Man or artificial intelligence? how to recognize machine-generated images and text
[ad_1]
The final solution envisaged by the company headed by the CEO Sam Altman could be a sort of invisible “watermark” to be glued to the contents to avoid using a ChatGPT-generated output by making it pass for human and consequently prevent the risk of plagiarism or someone’s writing style is malicious.
In the meantime, to recognize the authorship of a text or an image created by a bot, one can resort to software of various kinds (free or paid) and with different understanding capabilities.
The software list
The list of available solutions (including web interfaces, deep learning models and open source applications) is quite large but the fact remains that, in general, these tools (especially “free software”) are not able to identify all intelligences artificial potentially used to generate content.
The “classifier” (AI Text Classifier) developed by OpenAI and trained to detect “evidence” that a text was written by ChatGPT and the like, is a confirmation of this: by the startup’s own admission, the technology is constantly being refined and at the only partially reliable moment, opening up to the risk of “false positives”.
We are therefore talking about fallible software for recognizing the origin of a written content even if it is certainly useful for this purpose: GPTZero, for example, created by Edward Tian, a student at Princeton University, who analyzes a text starting from two main characteristics of the information reported in it, perplexity and burstiness.
[ad_2]
Source link
