The Washington Post has launched AI-generated podcasts, despite internal tests revealing significant errors and biases in the technology. An internal review indicated that over 68% of scripts from the feature “Your Personal Podcast” failed to meet the publication’s standards, with testers encouraged to err on the side of caution when assessing script quality. Failures included issues like mispronunciations and fabricated quotes, with the AI sometimes misrepresenting a source’s quotes as the paper’s stance. Despite these concerns, the product review team recommended proceeding with the release, acknowledging the tool as a work-in-progress prone to errors. Chief Technology Officer Vineet Khosla and Head of Product Bailey Kattleman emphasized the tool as aligning with the publication’s goals of enhancing customer experience through AI. The review concluded that minor adjustments were unlikely to improve script quality without increased risk, highlighting ongoing challenges in integrating AI with journalistic standards.
Source link
