This research examines AI models (OpenAI GPT-4 and Google Gemini 1.5 Flash) used to access Italian news across genres, focusing on the quality of information and sources (RQ1) and the models’ ability to provide correct links for users to verify and explore the news (RQ2). A custom GPT based on the Whittaker Grid serves as an experimental tool for evaluating the models’ sources. The debate on generative AI’s role in news reveals new biases from content-blocking by outlets, causing second-level distortions via copyright limits. Findings show robots.txt files widely restrict AI access, impacting reliability, while ChatGPT performances rate higher than Gemini ones.
Bad news: AI-LLMs and digital traces in journalism, the bias of copyright protection
Viviana CondorelliPrimo
;Fiorenza Beluzzi
Ultimo
2025-01-01
Abstract
This research examines AI models (OpenAI GPT-4 and Google Gemini 1.5 Flash) used to access Italian news across genres, focusing on the quality of information and sources (RQ1) and the models’ ability to provide correct links for users to verify and explore the news (RQ2). A custom GPT based on the Whittaker Grid serves as an experimental tool for evaluating the models’ sources. The debate on generative AI’s role in news reveals new biases from content-blocking by outlets, causing second-level distortions via copyright limits. Findings show robots.txt files widely restrict AI access, impacting reliability, while ChatGPT performances rate higher than Gemini ones.| File | Dimensione | Formato | |
|---|---|---|---|
|
Condorelli- Beluzzi- AIS 28-copia per autori.pdf
solo gestori archivio
Descrizione: Copia per gli autori
Tipologia:
Versione Editoriale (PDF)
Licenza:
NON PUBBLICO - Accesso privato/ristretto
Dimensione
315.69 kB
Formato
Adobe PDF
|
315.69 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
