Which AI Model Most Infringes on Copyrighted Content? « Machine Learning Times

Originally printed in AI Business, March 7, 2024. 
OpenAI’s GPT-4 reproduces essentially the most copyrighted content material from prompts amongst 4 common giant language fashions, based on new analysis from AI startup Patronus AI.
The startup, based by former Meta AI researchers, additionally discovered that common giant language fashions from the likes of Meta, Mistral and Anthropic generated copyrighted content material.
The startup examined OpenAI’s GPT-4, Anthropic’s Claude 2.1, Meta’s Llama 2 70B and Mistral’s Mixtral-8x7B-Instruct-v0.1.
GPT-4 reproduced copyrighted content material, on common, in 44% of prompts crafted to check how a mannequin regurgitates present content material. Mixtral-8x7B-Instruct-v0.1 produced copyrighted content material on 22% of take a look at prompts on common, whereas Llama 2 70B recreated content material on 10% of the prompts.
The mannequin that produced the bottom quantity of copyrighted content material was Anthropic’s Claude 2.1, with a mean rating of simply 8%.
To proceed studying this text, click on right here.


Recommended For You