When generative AI is discussed in relation to newsrooms, one typically thinks of the negatives, such as lawsuits or related issues with copyright, model training, or the threat of replacing reporters. However, some publications are embracing the technology, including The Washington Post, which launched its generative AI tool to optimize users’ content consumption.
Last week, The Washington Post debuted an experimental generative AI tool called “Ask The Post AI,” which allows users to get conversational answers on any topic referenced in text news articles published by the newspaper since 2016. The publication refers to the tools as an initiative “built by news for news.”
Also: I changed 5 ChatGPT settings and instantly became more productive – here’s how
The bot displays the results in order of relevant reporting; if none exist, it won’t populate anything. Then, a large language model will write a response to answer the question, although sometimes the answer may be pre-written.
“The changed search experience across the industry calls for us to meet the moment and meet audiences how, when, and where they want to be served with an updated user experience,” said Vineet Khosla, CTO for The Washington Post.
In the new feature’s FAQ section, The Post also explains that it has minimized the risk of hallucinations by training the AI exclusively on its own articles. Yet, the publication encourages users to cross-reference answers with the articles linked.
To get started, visit the site and ask any question, whether news-related or more general, such as “How can I stop junk mail?” Click on one of the pre-populated questions or type in your own question to test it out.
I asked, “What is the most populated city in the United States?” The results populated quickly, with two articles referenced, as well as a concise, conversational response that read, “The most populated city in the United States is New York City, with more than 8 million people.”
Neither the FAQ nor the announcement specifies who can access the tool; however, I was able to get the same results whether or not I was signed into my subscription. I could not click on the articles and read them without a subscription, which makes verifying the answer more challenging.