Article from twipemobile.com
At a time when texts generated by artificial intelligence (AI) are becoming increasingly common in newsrooms, there is an urgent need for verification and quality control. Such texts often contain inaccuracies or “hallucinations” – invented information that may seem convincing, but is untrue. To address this problem, the AI + Automation Lab at Bayerischer Rundfunk (BR) has developed Second Opinion – a content compliance check tool that helps journalists verify whether AI-generated summaries match the original text.
How does “Second Opinion” work?
Second Opinion acts as a second level of verification. Once a summary is generated by an AI model – such as ChatGPT – the tool compares it to the original text. Any discrepancies are highlighted: red for clear errors, and orange for partial inaccuracies. Users can click on these highlighted areas to see exactly where the problem lies – making the review process faster and more efficient for journalists.
While currently used primarily for automated summaries, this tool could also be applied to other types of texts, such as simplified versions for different audiences. However, it is important to note that Second Opinion does not perform fact-checking in the classic sense – it does not independently verify facts, but rather ensures that the AI-generated content matches the original source. For this reason, fact-checking still remains the responsibility of journalists and editors.
Open access to innovation
Part of the network AI4Media, BR has released Second Opinion as an open source tool to foster collaboration and innovation in the use of AI in media. Integrating this tool into editorial processes helps newsrooms increase the efficiency and accuracy of content using generative technology, without losing the essential role of human judgment.
In an increasingly complex media landscape, where technology offers quick but uncertain solutions, Second Opinion represents an important step towards balancing automation and reliability.