2023-12-15 16:10:00
Faced with the sudden arrival of generative artificial intelligence in the wake of ChatGPT, the media initially expressed an outcry. The industry has revolted once morest the supposed use of news articles to train AIs, without compensation. At the same time, she was extremely wary of the use of AI in editorial offices, considered too risky due in particular to hallucinations, these situations where the AI presents a lie as a truth. A double threat, in short.
But as the months go by, AI is becoming more and more common, and the media’s posture is starting to relax. Editorial staff are wondering regarding the relevant uses of new AI-powered tools, while on the other hand, OpenAI has just signed its second agreement with a press group to legally exploit its articles.
AI to support journalists
Sign of the times, Tuesday, the New York Times, often at the forefront of innovation in the sector through its influence and its human and financial resources, has appointed a director of AI initiatives. His role, among other things, will be to lead a small team within the editorial team of the illustrious journal, to conduct editorial experiments around generative AI.
For its part, the media specializing in tech and science Numerama last week deployed an AI summary feature for its articles, integrated into its new premium offer Numerama+. These summaries, point by point, echo a fairly widespread trend in the American tech press (Axios et Business Insidernotably) but also French (notably among 20 Minutes) since the mid-2010s.
« Even if it’s counterintuitive, these summaries are a vector of engagement. After reading the condensed information, readers become interested in the details, and this is reflected in the increased reading time spent on the article », summarizes Julien Cadot, director of editorial strategy at Numerama, supporting study.
Problem: this task proves not only boring for journalists, but also time-consuming. This is where AI comes into play. “To create content, AI makes too many mistakes. On the other hand, it works extremely well for summarizing text. We are therefore going to give him this task, which is very quickly tiring for journalists. », he explains.
Concretely, the media has added a new functionality in its text editor. Once the article is finished, the journalist clicks a button, which sends a request to GPT-4, OpenAI’s most powerful model. This button includes a custom-made and regularly refined prompt (a text given to the AI model), which triggers the generation of the summary. “ The prompt required real work to create, because it must work equally well on all our formats: tests, news, frequently asked questions, surveys, etc. », specifies Julien Cadot.
A few seconds later, GPT-4’s response is inserted in the dedicated space, and the author of the article only has to reread the small text, and correct it if necessary. “ AI-generated, verified by Numerama », Promotes the media. “ This task takes 30 seconds, while doing it by hand can easily take five minutes », summarizes the manager.
Axel Springer media arrives on ChatGPT
If AI begins to be integrated into the manufacturing of articles, the opposite is also true. On Wednesday, German press giant Axel Springer (Politico, Business Insider, Bild…) signed with great fanfare an unprecedented partnership with OpenAI, the company behind ChatGPT. According to the Financial Timesthe darling of AI will pay the publisher a license in the amount of “ eight digits “, in other words, more than ten million euros per year, over several years.
In exchange, ChatGPT will have access to all the group’s (paid) articles in near real time, and will be able to answer its users’ questions with short summaries of these articles. In other words, Axel Springer’s media are coming to ChatGPT! These responses will be accompanied by links to the full articles on the original media site, and the system will be deployed early next year.
The deal seems win-win. On the one hand, OpenAI addresses one of the main flaws of its tool, which has so far been very poor at providing truthful answers regarding recent events. For his part, Axel Springer offers unprecedented exposure to ChatGPT’s more than 100 million users per week, and a new audience or even subscription channel… in addition to being paid for! The agreement also grants OpenAI the right to train its AI on all articles in Axel Springer’s publications, including archives.
With the agreement, OpenAI also avoids a potential legal war. At present, the company, like its main competitors, does not disclose any details regarding the massive datasets on which its AIs are trained. But many rights holders, starting with the media, are aware that their content is being plundered by robots tasked with amassing as much text as possible.
OpenAI begins to defuse the media rumble
In September, several big names in the French press had made the decision to block the “ crawler » from OpenAI, the robot responsible for fetching texts on their sites. In other words, the issue of copyright has for several months been a veritable sword of Damocles that threatens to fall on the creators of AI models. The latter have the habit of pleading “ fair use “, that is to say reasonable use, when the question is addressed, while not specifying the contours.
But as noted Archyde.com, the IA Act approved in its non-final version on Friday December 8 might change the situation. As it stands, it provides for transparency obligations on the data used to train foundation models like GPT-4, of which ChatGPT is a version. That is, it might officially reveal that the AI model is trained on media articles.
With this non-exclusive agreement with Axel Springer, one of the most influential media groups in Europe, OpenAI is taking the lead in avoiding this disaster scenario. In July, he had already obtained an agreement with the Associated Pressin order to train its AI on the press agency’s archives.
« This is a great deal for both parties. But he opens the way to one-on-one negotiations, as in the case of neighboring rights », regrets Julien Cadot for his part. And to continue: “ The media, as an industry, has real power to leverage through its role in enriching AI. They might be the gold suppliers for jewelers who are the developers of AI models, and offer a global European license. But this type of sector agreement already seems unthinkable “. The risk ? Let the biggest media, capable of negotiating at a high level and waging complex legal battles, sign their agreements, and leave the smaller media, incapable of maintaining the balance of power, on the sidelines. OpenAI would thus minimize the legal risk, without having to pay for all rights holders.
1702695674
#outcry #media #begins #open #doors