『How USA Today fixed AI hallucinations in the newsroom | AI accuracy explained | The Signal Ep 13』のカバーアート

How USA Today fixed AI hallucinations in the newsroom | AI accuracy explained | The Signal Ep 13

How USA Today fixed AI hallucinations in the newsroom | AI accuracy explained | The Signal Ep 13

無料で聴く

ポッドキャストの詳細を見る

Welcome to The Signal, hosted by Fergus Bell and Tom Trewinnard. In this episode, Fergus speaks with Jessica Davis, Vice President of News Automation and AI Product at USA Today Network. Drawing on her extensive experience leading AI initiatives and capstone research at CUNY, Jessica addresses the critical challenge of scaling AI efficiently while preserving journalistic integrity.

She explains why constant manual oversight - having humans "in the loop" - creates operational bottlenecks and automation bias. Instead, she advocates for a "human on the loop" model, utilising structured AI evaluations and supervisory "evaluator" agents grounded in clear, data-informed metrics.

In this episode, Fergus and Jessica discuss:


  • Maintaining Trust: Why "generative AI is probabilistic" and how to manage accuracy risks.
  • Avoiding Oversight Fatigue: Moving past manual review without losing editorial control.
  • Evolving Standards & Hallucinations: How "evaluator agents" fix errors and prove reliability.
  • Embracing Industry Shifts: As Jessica notes: "I do think it's a natural step, and if it makes us uncomfortable, well, why? Let's sit with that a minute... and talk."


If you want to understand how the news is being created and distributed, how AI is - and can be - used in newsrooms, insights from media leaders, practical thinking on the future of journalism, and honest conversations about what's working and what isn't, subscribe and follow to stay close to The Signal.


Fergus and Tom will be here every week. Find The Signal on YouTube, Spotify, Apple and wherever you get your podcasts.


Produced by Rob Fitzpatrick

Edited by James Bullock

Hosted on Acast. See acast.com/privacy for more information.

adbl_web_anon_alc_button_suppression_t1
まだレビューはありません