Last week, when the Washington Post announced a new service that uses AI to generate podcasts, we immediately warned readers to brace for disaster.
Just as expected, the service — dubbed “Your Personal Podcast” — turned out to be an error-laden disaster right from the jump. As Semafor reported on Friday, many staffers at the embattled newspaper were outraged at the AI inventing and misattributing quotes and misinterpreting basic facts, just some of the extremely well-known and documented shortcomings of large language models.
Then the story got even worse. In followup reporting over the weekend, Semafor reported that WaPo leadership had launched the controversial feature despite knowing it was severely flawed. Internal tests found that the AI was generating podcasts riddled with errors and bias, with anywhere between 68 and 84 percent of the scripts deemed unpublishable by evaluators across three rounds of testing.
Despite the horrendous results, the company’s product review team pushed for the half-baked feature’s release, arguing it would “iterate through the remaining issues.”
It’s a perfect example of the troubled relationship between generative AI tech and journalism, from news writers publishing fake, AI-generated articles to Google discouraging users from visiting trustworthy news outlets by offering up error-laden AI summaries instead.
Not only does the intersection lead to a proliferation of mis- and disinformation, but experts have warned that it could further erode trust in once-reputable publications, including WaPo.
A spokesperson for the paper told Semafor that the feature is “currently in Beta” and that only if new features “prove to be successful for the customer do they then get launched.” In other words, there’s still a chance WaPo could give up on the idea.
But chances are that the Jeff Bezos-owned paper will continue to undermine its own journalism in the near future as part of a broader effort to cut costs. Earlier this year, WaPo launched an AI chatbot that was found to make plenty of mistakes. An AI writing coach dubbed Ember, which is meant to help write opinion pieces, is also in the works.
In the meantime, frustrated staffers at the paper are forced to watch as the chaos unfolds.
“It is truly astonishing that this was allowed to go forward at all,” a WaPo editor wrote in a Slack message following the latest AI podcast feature release, as quoted by Semafor. “Never would I have imagined that the Washington Post would deliberately warp its own journalism and then push these errors out to our audience at scale.”
... continue reading