mTags

LLMs (GPT, LaMDA…)Artificial intelligenceTools

Addition date

Jan 20, 2023 11:09 AM

Column

Raphael Thys

Link

https://medium.datadriveninvestor.com/gpt-zero-is-designed-to-recognize-texts-written-by-ai-ab7ff4d11fd6

LAN

EN

A student has developed an app that aims to distinguish AI -generated posts from texts written by humans

Photo by Hitesh Choudhary on Unsplash

Computer science student Edward Tian from Princeton University developed the GPT Zero app, which uses statistical characteristics to determine whether a text was written by a human or an AI.

According to Tian, ​​the app analyzes the text for even, constant complexity. This is a feature of AI-generated texts, while human writers are more likely to incorporate variations into their texts, he explained. The computer scientist published several case studies in which his algorithm was correct.

GPT Zero is currently heavily overcrowded because apparently many users want to test whether they have machine-generated texts or human-made works. GPT Zero is available on both the developer’s website and Streamlit.

We tried several times to test various texts created by Chat GPT, but the results were mediocre at best. For example, GPT Zero recognized longer texts created by Chat GPT for general, lexical knowledge with some degree of reliability. However, AI ​​recognition failed with short texts. According to the developer, at least ten sentences must be checked in succession to enable an evaluation.

With this experience in mind, it remains to be seen whether GPT Zero is actually a reliable way to recognize text created by AI. The developer wants to refine his model in the future. On the one hand, it should be borne in mind that the algorithms for recognition could be taken into account by future chatbots, and on the other hand, Chat GPT does not claim to always deliver correct answers.

Resistance to Chat GPT

The International Conference on Machine Learning (ICML) recently decided that authors should no longer use AI tools such as Chat GPT when publishing scientific articles. However, the ICML only prohibits texts “ completely produced “ by artificial intelligence. The organizers stressed that they do not ban the use of tools like Chat GPT “for editing or refining text written by authors”. The New York City Department of Education blocked access to Chat GPT on its networks over concerns that students were using the AI ​​tool.

Subscribe to DDIntel Here.

Visit our website here: https://www.datadriveninvestor.com

Join our network here: https://datadriveninvestor.com/collaborate