The Washington Post plans to continue publishing its personalized AI-generated podcasts despite internal criticism that the initiative flouts newsroom standards.
“This is how products are built and developed in the digital age: ideation, research, design and prototyping, development, then beta,” a Post spokesperson told TheWrap. “Only if they prove effective for the customer are they launched. As clearly stated on your personal podcast, it is currently in beta.”
The stance came after reports published Sunday shed more light on the product, which the Post launched last week as “the ultimate intersection” between “premium experiences, customer choice and AI platform and products.”
According to Semafor, about two-thirds of the product’s initial scripts failed a Post metric that determined whether the product met the newsroom’s standards. Log testers were asked to grade scripts on a pass/fail basis and, if in doubt, to fail the script as a precaution. On this curve, between 68% and 84% of scripts failed after three rounds of testing.
The Editorial Standards Slack channel also saw Post employees continue to speak out against the product. “I hope the product team understands that this isn’t just an overreaction from a bunch of retrograde Luddites,” one staffer wrote, according to Status. “These errors pose a threat to the core of what we do. »
Yet, according to Semafor, the product team believed they could “iterate” on any of the issues, as long as the editorial team labeled the tool as work in progress that could potentially generate issues. Among the most glaring flaws were fabricating quotes and offering comments that confused quotes with the journal’s position on a given issue.
Some of that frustration, which began last week, has been acknowledged internally. The newspaper’s standards manager, Karen Pensiero, took to a Slack channel last week to note that the episode was “frustrating for all of us.”
