In a statement, the news agency said OpenAI had access to an archive of Associated Press text stories dating back to 1985. In addition to licensing fees, the Associated Press will also get access to OpenAI’s technology, which it uses to experiment to determine how to improve journalism.
The news agency has used automation for years to produce some local sports reports and financial earnings reports. The Associated Press said it does not use “generative” technology (chatbots like ChatGPT) to write stories.
OpenAI, Google, and other AI companies have used billions of sentences extracted from the open internet to build “large-scale language models” that power chatbots. News articles, Wikipedia articles, social media comments, and blog posts have all been incorporated into the model without their owners’ permission, and tech companies generally claim public data is free to use.
A Washington Post analysis of the database of websites used to train one of OpenAI’s older AI models found AP’s primary news website to be the 68th most cited website in the database. It turns out that
An emerging group of authors, musicians, news outlets and social media companies argues that using their content to train AI would be a game-changer for how the internet works, especially since some AI tools are no longer human. It is repulsive because it is trained using . The content created is already being used instead of human workers. In the past two weeks, the industry has been hit by a wave of lawsuits alleging improper use of data, including a class action lawsuit against OpenAI and Google, and a lawsuit against comedian Sarah Silverman and two noted novelists against her OpenAI. I was.
The Washington Post reported Thursday that the Federal Trade Commission has launched an investigation into how OpenAI used consumer data to train its models.
“The dataset contains a lot of copyrighted content. The copyright owners have not consented to their use,” said Andres University of Miami law professor who studies intellectual property. Sawicki said. He said he could imagine tech companies and content creators signing more deals like the Associated Press to create a “clean database.”
“The problem is that the size of the dataset required to train the model is very large, so I think it would be very difficult to get consent from enough owners to make it technically viable. ‘ said Sawicky.
Chatbots like ChatGPT are trained on a set of information and cannot be continuously updated without retraining from scratch. In other words, it is not very useful for providing recent news or fresh information. Tech companies are trying to solve this problem by allowing chatbots to search the web themselves or ask questions from separate update databases. A contract with AP only allows OpenAI access to the archive, which is regularly updated with the latest news articles.
Tech companies have traditionally paid directly for news content for other uses. Both Google and Facebook pay news sites in some countries for direct access to the content they display on their platforms. In Australia, the government has passed legislation requiring this practice, and similar legislation is about to come into force in Canada.
