ChatGPT developer OpenAI said it is accelerating negotiations with dozens of publishers to reach agreements to license their exclusive content or articles for training on OpenAI data. This is a broader licensing effort than previously known becauseThe AI startup is looking for more content to train its artificial intelligence models, as well as gearing up for its "AppStore moment" - a custom GPT app store "GPTStore" based on its text-generating AI models such as GPT-4, which will launch next week.
"We are negotiating and discussing with a number of publishers. They are very active, they are very proactive, and it seems to me that it is going well." Tom Rubin, OpenAI's director of intellectual property and content, said in an interview with the media. "You've seen these deals announced and there will be more to come."
A person familiar with the matter previously told the media that OpenAI recently signed a multi-year licensing agreement with Axel Springer SE, the parent company of Politico, worth tens of millions of dollars. In July 2023, OpenAI announced an important agreement with The Associated Press (The Associated Press), but did not disclose the specific amount of the agreement. The deals are critical to OpenAI's future as the company balances the need for updated, accurate data to build models with increasingly strict scrutiny of data sources.
But last week, The New York Times, one of the media outlets that had negotiated with OpenAI, sued OpenAI and OpenAI’s majority shareholder Microsoft for using the magazine’s article without permission.
The lawsuit raises critical challenges to several of OpenAI's businesses. If The New York Times prevails, OpenAI could not only be owed billions of dollars, but it could also be forced to destroy all of its training data, including any data related to The New York Times content, an expensive and complex task. More immediately, however, the lawsuit complicates OpenAI’s efforts to strike collaboration deals with the media industry.
But Rubin said: "The current situation is very different from what publishers have faced in the past on search engines and social media. Here, content is used to train models, and it is not used to copy content, nor to replace original content."
However, the New York Times disagrees with OpenAI’s position and believes that ChatGPT plagiarized the work of its reporters without paying. In the lawsuit, the publisher showed examples of ChatGPT excerpting entire passages from the New York Times almost verbatim — and some reporters even pointed out that in some instances it specifically prompted ChatGPT to copy content from the New York Times. The publisher believes this proves that OpenAI illegally used data from The New York Times.
"If Microsoft and OpenAI want to use our results for commercial purposes, they are legally required to first obtain our permission," the New York Times said in a statement. "However, they have not done so."
If OpenAI fails to reach a cooperation agreement with the New York Times and more news content publishers, it will not only face huge compensation, but also face significant content loss due to data training, which will cause major obstacles to its many businesses, especially the company’s upcoming “GPT store”. The so-called GPTStore allows users to participate in the development of custom applications similar to ChatGPT and share and promote them. When others use the product, profits can be realized, which is equivalent to the "AppStore" based on OpenAI large model technology.
According to media reports, OpenAI plans to launch the GPT Store (GPTStore) sometime in the next week, which is a customized GPT application store based on its text generation AI model (such as GPT-4). It can be regarded as the "APPStore" launched by OpenAI combined with the GPT large model. GPTStore is designed to allow all users to share and sell GPT applications such as chatbots customized for different purposes and based on the GPT large model.
Launching a GPT application on the GPT store does not even require programming experience, so ordinary participants can also complete developing applications just like professional developers. Developers simply enter in simple language the functions they want their GPT application to provide, and OpenAI’s GPT application building tool GPTBuilder will attempt to create an AI-powered chatbot to perform those functions.
In the process of building GPT applications, developers only need to submit simple dialogue instructions, additional knowledge data, and then choose whether they need multi-modal functions such as network search, data analysis, and image generation. They can quickly develop similar ChatGPT applications in specific fields such as law, medical, and construction.
For example, by completing some simple pre-operations, a GPT application can be trained on a recipe set so that it can answer questions about specific recipe ingredients, and a GPT application focused on the recipe field is born. Alternatively, GPT could ingest a specific company's proprietary code base so that more specialized developers can inspect their application styles or generate code based on best practices.
OpenAI said developers building GPT applications must review the company's updated usage policy and GPT branding guidelines to ensure their GPT is compliant before they are eligible for public launch in the GPT store. They must also verify their user profile and ensure their GPT publishing is of a "public" nature.