AP gives OpenAI access to its vast news archive

OpenAI, a pioneer in the field of artificial intelligence (AI), has signed a license agreement with the Associated Press (AP) to use the AP's massive archive of textual stories to train its AI systems. Through this groundbreaking agreement, OpenAI will have access to the AP's text history archive, which dates back to 1985.

The length of the two-year agreement between OpenAI and AP is unknown, nor is the purchase price. As part of the agreement, the AP will provide OpenAI with access to its library of text stories that can be used to train its AI systems. In exchange, OpenAI will give the AP access to its technology, which it can use to test ways to improve the quality of its reports.

Some reports on local sports coverage and PA financial revenue have been auto-generated for years. However, "generative" technology (chatbots like ChatGPT) is not used by the AP in its reporting process.

There is growing controversy in the tech world over the use of copyrighted content for AI training. The "big language models" that power the chatbots of OpenAI, Google and other AI companies were trained on billions of sentences pulled from the public internet. These templates embed third-party content such as news articles, Wikipedia entries, social media comments, and blog posts without the authors' knowledge or consent.

A growing number of writers, musicians, media and social media companies are speaking out against the use of their work to train AI, saying it represents a fundamental shift in the nature of AI. Internet. They claim that artificial intelligence tools are already being used to replace humans by learning from human-made content. Since then, a slew of lawsuits have been filed against the industry, including class action lawsuits against OpenAI and Google and individual lawsuits filed against OpenAI by comedian Sarah Silverman and two prominent fiction writers.

The FTC has also launched an investigation into OpenAI's use of customer data for model training. The FTC requested documentation detailing OpenAI's efforts to mitigate potential security threats to its AI models and demanded an explanation of the company's efforts to improve the accuracy of its language models, which have turned out to be "mind-blowing" when asked a question to which they don't know the answer.

Because chatbots like ChatGPT are trained on a static set of data, they are not able to accommodate new information without being recycled. This makes them less reliable sources of up-to-date information and breaking news. To address this, some tech companies have systems in place that allow chatbots to perform their own web searches or get answers from a separate, constantly updated database. p>

The agreement between the two parties only grants OpenAI access to the AP archives, but these archives are regularly updated with the latest news. Tech companies have been known to pay for news articles in the past, but usually for a different purpose. Google and Facebook, in some countries, pay news outlets for the right to display their content directly on their platforms. The Australian government has passed legislation mandating the procedure, and a bill in Canada would do the same.

Both OpenAI and AP have expressed optimism about the development of AI, stating that they "believe in the responsible creation and use of these AI systems". Although the AP has used AI for nearly a decade in various capacities, including automating corporate earnings reports and summarizing certain sporting events, it said it currently does not use any generative AI in his reports.

He also recently launched an AI-powered image archive search and runs a program that helps local news organizations integrate AI into their operations. The Associated Press strongly supports a system that protects intellectual property and ensures authors are fairly compensated for their efforts.

Andres Sawicki, professor of intellectual property law at the University of Miami, said: "The datasets include a lot of copyrighted content. Rightsholders do not approve of these uses. It's not hard to engineer more agreements like AP's between tech companies and content producers with the goal of creating a "clean database. The problem is that the datasets needed to form the models are so large that I doubt it is possible to obtain permission from enough owners to make the technology practical.

In conclusion, the agreement between...