OpenAI GPT-4 Faces Allegations of Up to 44% Copyrighted Content Issues

Recently, Patronus AI, a company focused on evaluating large language models (LLMs), released a significant report examining the performance of four prominent AI models in dealing with copyrighted content. The findings revealed that OpenAI's GPT-4 model produced the highest percentage of prompts containing copyrighted text, accounting for an alarming 44%. This finding has sparked widespread discussion and concern within the industry about how AI models handle copyright issues.

In this evaluation, Patronus AI tested four models: OpenAI's GPT-4, Anthropic's Claude 2, Meta's Llama 2, and Mistral AI's Mixtral. Using the CopyrightCatcher tool, they analyzed the response of these models to prompts related to major copyrighted texts. Each model was tasked with generating content based on specific paragraphs or the opening segments of copyrighted books.

The results indicated that OpenAI's GPT-4 had the highest incidence of generating copyrighted content, at 44%. In contrast, Anthropic's Claude 2 showed a more cautious approach, with only 16% of prompts resulting in copyrighted material. This model also declined to respond to all requests about first paragraphs, citing the inability to access copyrighted material. Meta's Llama 2 produced copyrighted content in 10% of cases, while Mistral AI's Mixtral generated it in 6% of instances, spiking to 38% for first-paragraph prompts.

This report has ignited vigorous discussions among experts regarding copyright issues with AI models. They caution that these models could inadvertently infringe on copyright when generating text, especially with unlicensed materials. Therefore, it is crucial for content creators and businesses to ensure the legality and compliance of the content generated by AI.

Furthermore, the report urges copyright holders to stay informed about developments in AI technology and take proactive measures to protect their rights. This includes advocating for stronger copyright laws and clarifying the ownership and usage rights of AI-generated content. Patronus AI is committed to continuing its monitoring and evaluation of AI models concerning copyright issues and calls on relevant businesses and institutions to collaborate on ensuring the responsible development of AI technology.