OpenAI's GPT-4 Allegedly Contains Large Amount of Copyrighted Content, Up to 44%

Recently, Patronus, which specializes in evaluating large language models (LLMs) AIAn attention-grabbing report was released. The report states thatOpenAI(used form a nominal expression)GPT-4The model is filled with a large amount of copyrighted content, accounting for a whopping 44%.This finding has sparked a debate about theartificial intelligence (AI)Extensive discussion of the use of copyrighted material for models.

To test how well four mainstream AI models handle copyrighted content, Patronus AI used the CopyrightCatcher tool for analysis. They sent the same prompt words to GPT-4, Anthropic's Claude 2, Meta's Llama 2, and Mistral AI's Mixtral, asking for the output of a specified paragraph or the first paragraph of a copyrighted book.

However, the results were surprising. The GPT-4 generated the highest number of prompts with copyrighted text in the test, at 441 TP3 T. In contrast, Anthropic's Claude 2 showed the most caution in handling copyrighted content, generating only 161 TP3 T of completed prompts with copyrighted content. In addition, Claude 2 refused to answer all prompts for the first paragraph, citing unavailability of copyrighted material.

Meta's Llama 2 performed relatively well in dealing with copyrighted content, providing only 101 TP3T of copyrighted content in the cue. Mistral AI's Mixtral, on the other hand, provided 61 TP3T of copyrighted content, preferring to complete the first paragraph (381 TP3T).

This report has sparked controversy over the use of copyrighted material in AI models. On the one hand, there are concerns that these models may violate copyrights, leading to copyright disputes and legal issues. On the other hand, there are those who believe that AI models should be more careful in generating content and avoid using copyrighted content.

with regards toOpenAIThis report is certainly a challenge. They need to carefully consider how to protect copyright while ensuring that models such as GPT-4 do not infringe on others' copyrights when generating content. In addition, they will need to work with other organizations and copyright owners to explore effective ways to protect copyright in the field of AI.

This report reminds us of the copyright issues that may be involved in the generation of content by AI models. In the future, we need to work together to ensure that the development of AI technologies can take place in a way that respects and protects copyright.