NEW YORK, June 16 (Reuters) – Publisher Gannett is publishing stories as it and other news outlets begin to adopt popular technologies that could help them save costs and improve efficiency. The plan is to incorporate generative artificial intelligence into the system used for
But the largest US newspaper, with more than 200 daily newspapers, said it will involve humans in the process to prevent the technology from being automatically deployed without oversight. Renn Turiano, senior vice president and head of product at Gannett, said in a recent interview with Reuters that generative AI is a way to increase efficiency and eliminate some of the tedious tasks for journalists.
But Turiano didn’t mention any particular media outlets, adding that “for some of the other news services, the desire to get it out fast was a mistake.” “We are not making that mistake.”
Gannett isn’t alone in striking that balance. For example, Reuters President Paul Bascobert said in a statement on Thursday, in response to a reporter’s request for comment on the company’s plans, that the company has “a responsible approach that protects accuracy and fosters trust” in adopting AI technology. We are taking
Many US newsrooms are grappling with how best to incorporate AI tools that generate new content and data in response to user prompts and questions.
However, some experts say the limitations of generative AI, such as its tendency to “hallucinate” and provide false information under the guise of a veneer of certainty, are particularly important in industries where accuracy is a must. It is said to be a problem.
“At this time, we do not recommend these models for journalistic use cases such as automatic publishing to public channels,” said Nicholas Diakopoulos, an associate professor at Northwestern University.
Gannett’s strategy reflects the cautious approach taken by many mainstream news outlets.
Their warning comes in the wake of widely-reported generative AI gaffes in media such as CNET and Men’s Journal. Both publications used this technology to generate articles containing factual errors.
Next quarter, Gannett will roll out a live pilot program that uses AI to identify the most important points in articles and create bulleted summaries at the beginning of articles. The feature will be announced at USA Today in the fourth quarter. Journalists have the final say on whether to adopt content proposed by AI. Gannett will eventually incorporate its summarization technology into its publishing system.
Gannett’s journalists are struggling to keep technology from taking over. Hundreds were laid off on June 5, citing job cuts and stagnating wages. Generative AI has been part of the sticking point in negotiations with the company, the union said.
“Aside from replacing co-workers, the other concern is that they don’t think it’s an appropriate replacement,” said Ilana Keller, a journalist for the Asbury Park Press in New Jersey.
A spokesperson for the company said the use of AI is not meant to replace journalists, but as a tool to help journalists become more efficient and focus on creating more valuable content. .
Gannett laid off more than 600 employees last year after taking on $1.23 billion in debt from its merger with Gatehouse in 2019. However, the cost reductions have started to pay off.
“Not there yet”
As part of the push, Gannett is also developing a generative AI tool that splits long-form stories into different lengths and formats, such as photo bullet points and captions to create slideshows.
In a nutshell, Gannett relies on Cohere, a company that competes with Microsoft’s (MSFT.O)-backed OpenAI, which developed the ChatGPT chatbot. Gannett spent two weeks training Cohere’s large-scale language model based on his 1,000 previously published articles, including summaries written by reporters.
To further train the model, journalists from USA Today’s political team reviewed and edited the automated summaries and bulleted highlights.
While most news organizations have long relied on some form of artificial intelligence to recommend, personalize, and more content, new developments in generative AI are rekindling industry interest.
Gannett is also experimenting with natural language generation (NLG), a type of artificial intelligence that builds textual narratives and creates stories based on factual data. It doesn’t “think” like generative AI. Journalists review articles before publication.
Other news organizations are also approaching generative AI with varying levels of commitment and caution. The New York Times and Washington Post are in the planning stages, according to a Times memo obtained by Reuters and a statement by the Washington Post.
Bloomberg, which competes with Reuters, has developed its own generative AI model, called Bloomberg GPT, trained on financial data.
The New York Times, Washington Post and Bloomberg declined to comment further on the plan.
According to a May message to staff from Editor-in-Chief Alessandra, Reuters is using AI for voice-to-text transcription, for example to create scripts and subtitles for videos, but AI-generated articles, videos, The photos have not been made public. Galloni talks AI coaching for Reuters reporters.
BBC News Labs, an innovation incubator for broadcasters, is testing whether short-form description generation can be semi-automated.
For these articles, BBC News Labs took pre-published pieces of BBC content and built prototypes using the ChatGPT-3 model. BBC News Lab head Miranda Marcus said: “Unless a journalist manually pulls it out, it can never get anywhere near the viewer.”
“There’s a whole other world in terms of what kinds of stories you can tell with these tools,” Marcus added. “But we’re not there yet.”
Reported by Helen Koster.Editing: Anna Driver
Our standards: Thomson Reuters Trust Principles.
