The good, the bad, and the completely made-up: Newsrooms on wrestling accurate answers out of AI
Briefly

Erlend Ofte Arntsen, a journalist, has submitted many Freedom of Information Act requests, accumulating triple to quadruple digits when related inquiries are counted. A new assistant, FOIA Bot, utilizes generative AI to streamline his workflow, addressing responses from the government and saving valuable time. This technology, based on retrieval-augmented generation (RAG), ensures high reliability by pulling information from a specific database instead of general sources, enhancing the accuracy of generated responses. Arntsen finds the bot a substantial aid in managing requests while focusing on investigative reporting amid tight deadlines.
Erlend Ofte Arntsen has filed more Freedom of Information Act requests than he can count - triple digits by one tally, quadruple when you include follow-ups and related requests.
FOIA Bot is part of an emerging tech stack of newsroom tools that leverage a specialized AI architecture called retrieval-augmented generation, or RAG.
When the government sends back a request or rejection, the bot comes up with a competent rejoinder, given its access to the whole of Norway's FOIA law and 75 templates of similar responses.
A RAG-powered model retrieves information from a journalist-defined database, then uses that to augment what it generates with attributions to boot.
Read at Nieman Lab
[
|
]