This new technology from OpenAI promises to detect AI texts in student work, no more cheating?

[ad_1]

OpenAI is preparing a tool that can spot texts written by ChatGPT. This innovation could help professors in checking students’ work. However, concerns about its potential impact are holding back its implementation.

chatgpt
Source: 123rf

Last year, OpenAI claimed that it was impossible to detect whether a text had been written by ChatGPT. For the start of the school year, the company even distributed a guide to teachers on how to use AI in education, rather than punishment. However, the mass adoption of artificial intelligence by students changed the situation and forced the company to reconsider its position.

Faced with this situation, OpenAI has developed a tool to detect texts generated by ChatGPT. The latter uses a technique of watermark which consists of making small invisible adjustments in the selection of words that AI uses. These subtle modifications thus make it possible to recognize whether a text has been written by its artificial intelligence.

OpenAI Hesitant to Deploy Its ChatGPT-Generated Text Detection Tool

The watermarking method developed by OpenAI stands out for its precision. According to a company spokesperson, this technique is effective against local falsification, such as partial rewording of a text. However, it has limitations facing more complex rewritessuch as using translation systems or modifying with another AI model. This tool could provide a solution for teachers, who are largely overwhelmed by its growing use. A recent study revealed that 99% of students use generative AIs like ChatGPT, a staggering figure that testifies to the scale of the phenomenon.

However, OpenAI is aware of the implications of this technology. The company fears that making this tool available could deter some users to use ChatGPT. Additionally, researchers have expressed concerns about its potential impact on non-English speakers. They could be unfairly penalized by text watermarking due to language differences and the way AI handles languages ​​other than English. The company is currently still evaluating the benefits and risks of the tool before deciding whether to make it more widely available.

Source: The Wall Street Journal

[ad_2]

Source link -101